Opportunistic Networks

Opportunistic Networks
Fundamentals, Applications and Emerging
Trends

Edited by Anshul Verma, Pradeepika Verma,

Sanjay Kumar Dhurandher and Isaac Woungang
First edition published 2022
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-­2742
and by CRC Press
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
CRC Press is an imprint of Taylor & Francis Group, LLC
© 2022 Taylor & Francis Group, LLC
The right of Anshul Verma, Pradeepika Verma, Sanjay Kumar Dhurandher and Isaac
Woungang to be identified as authors of this work has been asserted by them in
accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
Reasonable efforts have been made to publish reliable data and information, but the
author and publisher cannot assume responsibility for the validity of all materials or
the consequences of their use. The authors and publishers have attempted to trace
the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we
may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted,
reproduced, transmitted, or utilized in any form by any electronic, mechanical, or
other means, now known or hereafter invented, including photocopying, microfilming,
and recording, or in any information storage or retrieval system, without written
permission from the publishers.
For permission to photocopy or use material electronically from this work, access
www.copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-­750-­8400. For works that are not
available on CCC please contact [email protected]
Trademark notice: Product or corporate names may be trademarks or registered
trademarks and are used only for identification and explanation without intent to
infringe.
ISBN: 978-­0-­367-­67730-­5 (hbk)
ISBN: 978-­0-­367-­67732-­9 (pbk)
ISBN: 978-­1-­003-­13258-­5 (ebk)
Typeset in Sabon
by Straive, India
Contents

Preface vii
Acknowledgments ix
Contributors xi

1 Mobile-­Code-­Based Opportunistic Networking 1

CARLOS BORREGO IGLESIAS

2 Opportunistic Emergency Scenarios: An Opportunistic Distributed

Computing Approach 21
CARLOS BORREGO IGLESIAS

3 Reactive and Proactive Routing Strategies in Mobile Ad Hoc Network 37

MD. IBRAHIM TALUKDAR AND MD. SHARIF HOSSEN

4 Secure Hierarchical Infrastructure-­Based Privacy Preservation

Authentication Scheme in Vehicular Ad Hoc Networks 55
MARIYA OUAISSA, MARIYAM OUAISSA AND MERIEM HOUMER

5 Simulation Tools for Opportunistic Networks: How to Set Up and

Simulate the ONE Simulator 81
MD. SHARIF HOSSEN

6 Understanding Influencers of Adaptive Social-­Aware Opportunistic

Forwarding 103
SOUMAIA A. AL AYYAT AND SHERIF G. ALY

7 Performance Analysis of AODV and DSDV Routing Protocols in Mobile

Ad Hoc Network Using OMNeT++ 141
RUKHSANA NAZNIN AND MD. SHARIF HOSSEN

v
vi Contents

8 Message Forwarding and Relay Selection Strategies in Mobile

Opportunistic Networks 155
MOHD YASEEN MIR AND CHIH-­LIN HU

9 Routing Techniques for Opportunistic Network 173

SONAM KUMARI AND ITU SNIGDH

10 Blockchain Leveraged Node Incentivization in Cooperation-­Based Delay

Tolerant Networks 187
SOUVIK BASU, SOUMYADIP CHOWDHURY AND SIULI ROY

11 Evaluation of Energy Efficiency of Opportunistic Network

Routing Protocols 207
MD. KHALID MAHBUB KHAN, MUHAMMAD SAJJADUR RAHIM AND
ABU ZAFOR MD. TOUHIDUL ISLAM

12 Mobility Models in Opportunistic Networks 225

JAGDEEP SINGH, SANJAY KUMAR DHURANDHER AND VINESH KUMAR

13 Opportunistic Routing in Mobile Networks 243

ZEHUA WANG, YUANZHU CHEN AND CHENG LI

14 Security in Opportunistic Networks 299

MOHINI SINGH, ANSHUL VERMA AND PRADEEPIKA VERMA

Index 313
Preface

OVERVIEW AND GOALS

The opportunistic network is a result of intensive research of so many years on Mobile ad

hoc networks (MANETs), therefore it is called an extension of MANET. Like MANET, an
opportunistic network is also an autonomous, dynamic, decentralized, and infrastructure-­
less network that is made of wireless mobile devices. It is also ad hoc because the network
is formed dynamically “on the fly” and does not have the pre-­existing infrastructure, unlike
infrastructure-­full networks where routers and access points are used to manage network
information. Instead, each node acts as an end node (source or destination) as well as a router
(forward data of other nodes). The next forwarder of the data is determined dynamically
based on network connectivity and the routing protocol in use. Unlike MANETs, opportu-
nistic networks consider several wireless networks’ properties, such as nodes mobility, dis-
connection of nodes, network partitions, and links’ instability, as advantages rather than
exceptions. This often results in the design of opportunistic networks being significantly
more complex and challenging than that of MANETs.
In opportunistic networks, mobility of the nodes is used as a way for the communication
between disconnected clusters of nodes rather than an exception that should be overcome.
It does not create a complete path between a source and destination during a transmission.
There is no need for source and destination to be connected to the same network at the same
time. Despite this, messages can be exchanged between source and destination by using the
store-­carry-­and-­forward approach in which intermediate nodes store messages in the absence
of better forwarding nodes, carry messages from one location to other, and forward to the
other nodes those that have better opportunity to bring the messages closer to the destina-
tion. This approach of message transmission in highly intermittent connected ad hoc wireless
networks became more favorable than the traditional MANET paradigm.
However, the opportunistic network is an emerging and recent area of research, and
many industrial applications have been developed based on this network. The opportunistic
network is made of heterogeneous mobile devices that exist everywhere, due to the widespread
use of cellular and pervasive mobile devices by the people, and are emerging day by day.
Hence, many more concrete futuristic mobile applications can be developed by adopting the
opportunistic networking paradigm. To make this research area more adaptable for practical
and industrial use, there is a need to further investigate several research challenges in all
aspects of opportunistic networks. Therefore, this book will provide theoretical, algorithmic,
simulation, and implementation-­ based research developments related to fundamentals,
applications, and emerging research trends in opportunistic networks.
The following are some of the important features of this book, which, we believe, would
make it a valuable resource for our readers:

vii
viii Preface

• This book is designed, in structure and content, with the intention of making the book
useful at all learning levels.
• Most of the chapters of the book are authored by prominent academicians, researchers,
and practitioners, with solid experience and exposure to research on the area of routing
in opportunistic networks. These contributors have been working in this area for many
years and have a thorough understanding of the concepts and practical applications.
• The authors of this book are distributed in a large number of countries, and most of
them are affiliated with institutions of worldwide reputation. This gives this book an
international flavor.
• The authors of each chapter have attempted to provide a comprehensive bibliography
section, which should greatly help interested readers to dig further into the aforemen-
tioned research area.
• Throughout the chapters of this book, most of the core research topics have been cov-
ered, making the book particularly useful for industry practitioners working directly
with the practical aspects behind enabling the technologies in the field.

We have attempted to make the different chapters of the book look as coherent and synchro-
nized as possible.

TARGET AUDIENCE

This book will be beneficial for academicians, researchers, developers, engineers, and practi-
tioners working in or interested in the fields related to opportunistic networks, delay tolerant
networks, intermittently connected ad hoc networks, vehicular ad hoc networks, and mobile
ad hoc networks. This book is expected to serve as a reference book for developers and engi-
neers working in the mobile application development and telecommunications industries,
and for a graduate/postgraduate course in computer science and engineering/information
technology.
Acknowledgments

We are extremely thankful to the 28 authors of the 14 chapters of this book, who have
worked very hard to bring this unique resource forward for helping the students, researchers,
and community practitioners. We feel that it is contextual to mention that as the individual
chapters of this book are written by different authors, the responsibility of the contents of
each of the chapters lies with the concerned authors.
We like to thank Randi Cohen, Publisher – Computer Science & IT, and Gabriella Williams,
Editor, who worked with us ohe beginning, for their professionalism. We also thank Daniel
Kershaw, Editorial Assistant at Taylor & Francis Group, and his team members who tire-
lessly worked with us and helped us in the publication process.
This book is a part of the research work funded by “Seed Grant to Faculty Members under
IoE Scheme (under Dev. Scheme No. 6031)”1 and “DST-­Science and Engineering Research
Board (SERB), Government of India (File no. PDF/2020/001646)”2.

NOTES

1 Awarded to Anshul Verma at Banaras Hindu University, Varanasi, India.

2 Awarded to Pradeepika Verma at Indian Institute of Technology (BHU), Varanasi, India.

ix
Contributors

Soumaia A. Al Ayyat Abu Zafor Md. Touhidul Islam

The American University in Cairo University of Rajshahi
New Cairo, Egypt Rajshahi, Bangladesh

Sherif G. Aly Md. Khalid Mahbub Khan

The American University in Cairo University of Rajshahi
New Cairo, Egypt Rajshahi, Bangladesh

Souvik Basu Vinesh Kumar

Heritage Institute of Technology Bharati College, University of Delhi
Kolkata, India New Delhi, India

Yuanzhu Chen Sonam Kumari

Memorial University of Newfoundland Raj Kumar Goel Institute of Technology
St. John’s, Canada Ghaziabad, India

Soumyadip Chowdhury Cheng Li

University of Engineering and Management Memorial University of Newfoundland
Kolkata, India Canada

Md. Sharif Hossen Mohd Yaseen Mir

Comilla University National Central University
Cumilla, Bangladesh Taoyuan, Taiwan

Meriem Houmer Rukhsana Naznin

Moulay Ismail University Comilla University
Meknes, Morocco Cumilla, Bangladesh

Chih-­Lin Hu Mariya Ouaissa

National Central University Moulay Ismail University
Taoyuan, Taiwan Meknes, Morocco

Carlos Borrego Iglesias Mariyam Ouaissa

Universitat de Barcelona Moulay Ismail University
Barcelona, Spain Meknes, Morocco

xi
xii Contributors

Muhammad Sajjadur Rahim Md. Ibrahim Talukdar

University of Rajshahi Comilla University
Rajshahi, Bangladesh Cumilla, Bangladesh

Siuli Roy Anshul Verma

Heritage Institute of Technology Banaras Hindu University
Kolkata, India Varanasi, India

Jagdeep Singh
Sant Longowal Institute of Engineering Pradeepika Verma
and Technology, Longowal Indian Institute of Technology (BHU)
Sangrur, India Varanasi, India

Mohini Singh Zehua Wang

Banaras Hindu University University of British Columbia
Varanasi, India Vancouver, Canada

Itu Snigdh
Birla Institute of Technology Sanjay Kumar Dhurandher
Mesra, India Netaji Subhas University of Technology
New Delhi, India
Chapter 1

Mobile-Code-Based Opportunistic
Networking
Carlos Borrego Iglesias
Universitat de Barcelona, Barcelona, Spain

CONTENTS

 1.1 Introduction............................................................................................................... 1
 1.2 Motivation................................................................................................................. 2
1.2.1 IP Multicast: A Case Example ......................................................................... 3
  1.3 Scenario Description................................................................................................. 4
  1.4 Active Messages........................................................................................................ 6
  1.5 General Protocol Architecture................................................................................... 7
  1.6 System Architecture Elements ................................................................................... 8
  1.7 Security Considerations............................................................................................ 11
 1.8 Reification................................................................................................................ 12
  1.9 Primitive Services Types and Task Delegation........................................................... 14
1.10 Dynamic Multi-Routing........................................................................................... 15
1.11 Aggregation, Scheduling, and Dropping................................................................... 16
1.12 Application Influenced Movement Model ...............................................................17
1.13 Active Messages: A Distributed Sensing Infrastructure............................................. 18
References.......................................................................................................................... 19

1.1 INTRODUCTION

The Internet is the evidence of how flexible and sturdy the TCP/IP network architecture has
proved to be to create a global network of networks. Although there are many important
issues about this architecture, there are also a good number or reasons to commend it. But
there are some networks for which the TCP/IP is not an option because of, among other
things, its relatively short timeouts or its routing algorithm. This is the case, for instance,
when several devices are able to communicate to each other only once in a while, being
unreachable most of the time. The nodes in these types of networks could use their peers to
asynchronously forward information for them from node to node to any point of the net-
work, even if this process took a long time waiting for the opportunity to transmit the infor-
mation. This is not a new idea at all, and many architectures in the literature provide similar
solutions, usually under the labels of Challenged Networks, Opportunistic Networking, or
Delay and Disruption Tolerant Networking (DTN).
In these networks, a critical issue, if not the most important, to deal with is routing. Because
there is not a fixed topography of the network, the decision-making process to choose the
next hop among the neighbors cannot rely on it. Instead, routing has to be a dynamic function

1
2  Opportunistic Networks

based on the local information acquired by the router in the very moment of the routing pro-
cess. Influenced by the success of the Internet, full of general purpose mechanisms, one could
think that a holistic “one-fits-all” routing algorithm would be useful here and would estab-
lish a standard for opportunistic networks. Unfortunately, the approaches trying to do that
are bound to fail. Such a standard, should it be, would lead to deadlocks similar to the ones
faced in the current Internet (e.g. multicast routing), because not all the applications have
the same requirements and not all opportunistic networks are alike. Different applications
may have different goals when it comes to routing the information, such as minimizing the
latency time, minimizing the variation in latency time, maximizing the throughput, maxi-
mizing the reliability, or minimizing the routing overhead. In fact, most of the successful
solutions for doing the routing in these opportunistic networks fit just a particular scenario,
normally consisting of a single application [1], and they set the trend. Some of these works
claim to provide a generic framework for any type of scenario/application, but at the end of
the day they fail to include new routing policies once the network has been deployed. On
the other hand, it would be very convenient to let the applications decide the routing instead
of trying to design a common algorithm including the associated fundamental properties for
all scenarios and applications. This is definitely the way of optimizing the performance of
this type of networks: there is no better source for routing information than the applications
themselves.
In this chapter, we introduce a general purpose architecture for DTN based upon the idea
of letting the application, by means of its messages, decide the routing that will take place
in every node. The keystone of this proposal is carrying the routing algorithm code along
with every single message. The resulting DTN can be used by different applications, even if
they were not foreseen before the deployment of the network. Thus, the DTN is no longer
bound to specific applications and becomes a real, open, general-purpose heterogeneous
network as it has been on the Internet for connected networks. A network built following this
architecture is flexible and open, and application designers can make the most appropriate
use of it according to the specific requirements of the application. In short, the evident
dependence between an application and its routing protocol is preserved while keeping other
applications’ interests.
This chapter presents a general architecture for opportunistic networks based on mobile
code. A formal definition of all the actors involved in communication networks is presented.
A generalization of these communications is described making our proposal a general case
of communications information exchange. Several DTN issues such as routing, DTN conges-
tion and error handling are discussed.

1.2 MOTIVATION

We have seen in the previous section the existence of different applications which run on
scenarios where there are intermittent connectivity, asymmetric bandwidths, long and vari-
able latency or ambiguous mobility patterns. In these scenarios, the network is created when
mobile and/or static nodes holding wireless devices intermittently connect. These network
limitations are due to issues such as small wireless radio range, mobility of the nodes, energy
constraints, network attacks or noise. Classic network protocols, such as TCP/IP, as explained
in previous sections, do not solve these challenging issues. For these purposes different pro-
posals such as the Bundle Protocol [2] and Haggle [3] have been floated. These proposals
evolve around the store-carry-and-forward paradigm: intermediate mobile or static nodes
accept the custody application information until they are able to find new custodians to for-
ward this information.
 Mobile-Code-Based Opportunistic Networking  3

For every different application which is intended to be run on a DTN scenario, we get
from the mentioned proposals, different interesting solutions which include various ways
of solving routing problems. These proposals include replication-based protocols, such as
epidemic and PRoPHET [1]; forwarding-based protocols, such as MEED [1] and LAROD
[1]; and a great many others. We believe that these proposals have been extremely useful to
understand the complexity of the problematic issues around DTN networks. However, these
different proposals are studied separately, defining a custom DTN network for every different
scenario.
We propose a step further in DTN networks. We believe in the necessity of a common
network in which different applications may coexist without leaving behind their optimal
routing algorithms. Traditional bundle-based proposals [2] allow just one type of routing
algorithms. When sharing a unique DTN infrastructure among several applications, a unique
routing algorithm must be chosen. We believe this is not efficient and can be improved. Our
proposal intends to break this tightly coupled relationship among DTN infrastructures and
the applications.
Contrary to well-established protocols like TCP/IP, a unique routing algorithm for many
applications has proven to work efficiently, and the Internet is the most compelling proof of
it. However, there are different applications in the context of DTN networks, which work
optimally just when using different routing algorithms. If we want to let these types of differ-
ent applications share a unique network infrastructure we need to propose a new paradigm.
This paradigm will be described in this chapter.

1.2.1 IP Multicast: A Case Example

Multicasting, in the context of TCP/IP networks and concretely in IPv4, is the sending
of information over a TCP/IP internet to multiple destinations with a single datagram.
Different applications such as software distribution, multimedia streaming, mobile TV, radio
broadcasting, and new media services, with different needs, take advantage of Multicast in
order to spread application data all around the Internet.
Datagrams sent to Multicast groups have as their destination IP addresses an Internet
address belonging to a special range of addresses called D type. Internet hosts subscribe or
unsubscribe dynamically to these Multicast groups to receive or stop receiving application
data from the application source.
The network infrastructure is responsible for replicating the Multicast datagrams and
intelligently routing these according to the topology of receivers interested in that informa-
tion, that is, Internet hosts who are subscribed. On one hand, hosts advertise to their local
routers about their interest to belong to a given Multicast group, establishing the Multicast
group membership. On the other hand, several different protocols have been defined to com-
municate among local and remote Multicast routers in order to route Multicast traffic from
the Multicast source application to the many Multicast clients. Examples on these protocols
can be found in [4] and include protocols such as PIM Sparse Mode, PIM Dense Mode, PIM
source-specific Multicast, and Bidirectional PIM. These protocols are widely deployed all
around the Internet.
Applications willing to use Multicast use the different network infrastructures using the very
same protocols to route Multicast datagrams. Unfortunately, no general-purpose Multicast
routing algorithm covering every router on the Internet is available. This issue forces sender
applications to choose from the different protocols in terms of the better adjustment of
their needs and access to the network infrastructure. For example, service providers offer
multimedia services to their clients using their network infrastructure which employ optimal
Multicast routing protocols chosen in terms of the type of multicasted applications, but do
4  Opportunistic Networks

not work well for other types of applications. These networks remain isolated for other
applications willing to Multicast since different PIM protocols do not understand each other.
As a result, no global Multicast service is available on the Internet. Unfortunately, Multicast
is not the only domain in which the coexistence of different protocols is not entirely solved.
In the context of mobile ad hoc networks (MANETs), different routing protocols have been
proposed, but when it comes to support heterogeneous MANETs, no common routing para-
digm copes with all the desired requirements. Our proposal in this chapter intends to avoid
precisely these types of failures in DTN scenarios.

1.3 SCENARIO DESCRIPTION

DTN research tries to provide solutions to problems where network partitions are created
due to intermittent connectivity by using the already-described store-carry-and-forward
switching. We propose in this study an enhanced version of this paradigm based on mobile
code. In this section, we first introduce the different actors and functions involved in DTN
scenarios. Following this, a six-phase general case of protocol is described.

Application list: We define applist as a non-empty set {app1, app2, …, appn} as the list of n
applications present in a network scenario. Applications mainly create messages which
are sent over the network from a source to a destination.
Custodian list: Let {c1, c2, …cc} be the list of c nodes capable of temporally storing applica-
tion messages, carrying them, and routing them to some other custodian nodes.
Application messages: Let {messappi,1, messappi,2, …, messappi,m, } be the list of m messages an
application appi sends over the network. Let messagesc be the list of messages a custo-
dian node c has at a given time.
Custodian routing information tree (RIT). For every custodian node cc, routing information
may be stored locally in the custodian nodes. A RIT may be defined as a root node,
RITroot or t + 1 number of RIT trees {RITlocal, RITapp1, RITapp2, …, RITappt} where t is the
number of the different applications present in the network. These trees contain other
RIT trees or values for routing purposes.
Custodian buffer space. The amount of space a custodian node ci has available for
application messages, and routing information is defined as buff(ci).
Custodian movement model: Custodian nodes follow movement models mcj which
affect important network parameters such as connection window time and encounter
statistical distribution.
Application custodian code. Let codeapp,cj be an application code an application app
intends to execute on a custodian node cj.
Neighbor list: We define neighbourList(cj, t), as the list of p neighbors {n1, n2, …np} a
custodian node c has at a given time t.
Routing function: Given an application message messapp,i, belonging to application app, a
custodian node cj, and its local RIT tree RITcj, let frouting(messapp, i, neighbourList(cj, t), RITc)
be the function which returns a subset of the neighbors of the custodian node cj which
should custody the application message.

In general, network protocols follow three traditional independent phases: Data creation
phase, routing and delivery phase. Besides the above-mentioned store-carry-and-forward
paradigm, we define the following communication paradigm phases (Figure 1.1):
 Mobile-Code-Based Opportunistic Networking  5

(a) (b)

(c) (d)

(e) (f)

Figure 1.1  (a) Application message creation phase: In this phase, application app creates messapp, i. (b) Store phase:
Message messapp, i is stored at buffer (custodian c) in custodian custodian c. (c) Carry phase: The mes-
sage follows the custodian movement(custodian c) movement model. (d) Process phase:The custodian
node cj executes local code codeappa, cj on behalf of application appa. (e) Routing phase: The message is
replicated following its routing algorithm, which in this example returns node2. (f) Delivery phase:The
message messappa, i is freed from the last custodian buffer (buff (cj)) and delivered to the application
layer on custodian cj.

Data creation phase. We define data creation for application app as a process in which
application app creates m application messages messapp,i, i in [0..m]. Application mes-
sages may be created as the result of application programs or application custodian
codes (codeapp,cj). The format of data messages will be described in Section 1.4.
Store phase: Messages are stored in a custodian node cj using the buff(cj). Since this space
is limited, the routing problem can be considered as an optimal resource allocation
challenge.
Carry phase: Messages are carried from one place to another. Messages perform active
carries, that is, carrying application app information from custodian node to custodian
node in terms of the function frouting(messapp,i, neighbourList(cj, t), RITcj). Instead, a cus-
todian node if in motion, performs a passive carry when moving. In this case, there is
no software entity performing the transport itself [5]. Instead, there is a physical move-
ment of the custodian node m(cc) who performs the carry action.
Process phase: A custodian node cj may execute codeappi, cj on behalf of the application app.
For these purposes, a custodian node needs to have execution environments. This phase
is equivalent to the process phase in the store-carry-process-and-forward paradigm that
will be explained in Chapter 2.
Forwarding phase: This phase occurs given custodian node cj and application information
message messapp,i belonging to application app. This phase is equivalent to the forward
phase in the already defined store-carry-and-forward paradigm. Message messapp,i is
propagated in the network following the following expression. Note that if forward-
ToList = Ø, it means that the message is discarded.


forwardToList  frouting messappi , neighbourList  c j , t  , RITcj , where
forwardToList  neighbourList  c j , t   c j
6  Opportunistic Networks

Delivery phase: Once a message messappa, i has arrived to its destination, it is passed to the
destination application and removed from the last custodian buffer (buff (cdestination)).
Other copies of message messappa, i may be present in the network in other custodian
nodes.

We propose the idea of creating a network to allow different applications coexist without
leaving behind their optimal routing algorithms. Our proposal is leaving this routing decision
to the very same application. This will let the different applications determine which routing
algorithm to apply in every DTN hop. In order to achieve this flexibility, two options must
be studied.
The first option involves a deployment of all of the possible routing protocols frouting(messap
pa,i, neighbourList(cj, t), RITcj) in the different DTN custodian nodes {c1, c2, …, cm}. This option
presents some immediate disadvantages. Routing algorithm deployment is expensive in chal-
lenged networks and it is not guaranteed that this deployment of all the necessary frouting
functions from the different applications will arrive at every custodian node. Furthermore,
newcomer applications will need to make additional routing deployments.
Second, we could let the very same application message mess carry along with itself the
routing algorithm frouting(messappa, i, neighbourList(cj, t), RITcj) and/or the application custodian
code, codeappa, cj. Theses algorithms must be able to be run in any custodian node. Mobile
code is a good candidate to implement a data-driven approach to DTN networks.
In some scenarios with limited storage space, having replicated the same routing algorithm
code on every message belonging to the same application could be inefficient. Every time the
active message is forwarded, besides message data, the routing code must be sent to the des-
tination platform. These static codes can be cached on the different node execution environ-
ments, using as described in [6], a global cache service to efficiently deal with the distribution
of agent codes. Caching these codes improves and speeds up message forwarding.

1.4 ACTIVE MESSAGES

Messages containing application data, routing algorithms frouting(messappa,i, neighbourList

(cj, t), RITcj) and application codes codeappa, cj are called active messages. In Figure 1.2 the four
fields that define an active message are depicted. These fields are:

Delivery information: In this field the source and destination addresses, active message
creation timestamp, and static lifetime is defined.
Application code: The application code codeapp, cj creates and manipulates application
data. This data may be created beyond the source custodian node. Data belonging to
an active message is reviewed every time the application code is executed on a custo-
dian node.
Routing code: Routing algorithm code frouting(appa) chooses from the available neigh-
bors on the list of the potential information custodian nodes. The routing code will be
responsible for defining the path the application data will take.
Data payload: Application data created by the source application.

Figure 1.2  Active message fields: control information, routing algorithm, and data payload.
 Mobile-Code-Based Opportunistic Networking  7

Figure 1.3  Active message forwarding. Custodian Node A forwards an active message containing some data, its
routing code, and an application code to Custodian Node B.

We propose a simple paradigm, depicted in Figure 1.3, in which the application information
is carried by the active messages. Active messages can be seen therefore as a transmission
medium, i.e., each active message is functionally analogous, conceptually speaking, to a car-
rier pigeon, as in [7]. DTN custodian nodes must include an execution environment which
implements the DTN layer that handles delays and disruptions. The application code creates
or modifies application data which is encapsulated inside the active messages. Active mes-
sages stay in the custodian nodes, which accept their custody, until they are able to be for-
warded to another one or they arrive at their destination node. Custodian nodes may employ
different convergence layers, the interfaces between the common Bundle Protocol and the
specific inter-network protocol suites.

1.5 GENERAL PROTOCOL ARCHITECTURE

Layering protocols is an excellent way to simplify networking designs by the means of

separating the latter into functional layers, and defining protocols to delegate each layer’s
task. In the context of network architectures based on layers, every layer receives from the
previous layer its protocol data unit. From the moment of this reception, this protocol data
unit becomes the service data unit of the new layer, which upon adding a header from the
new layer, a protocol data unit of the new layer is obtained. Each layer interacts, mainly
with its neighboring layers, across the interfaces between them. This mechanism is called
encapsulation and guarantees the layer principle in which different layers are independent
from one another. Traditional de facto standard protocols like TCP/IP break this layer
principle for practical reasons, letting some information from one layer be employed by its
following layer.
On the other hand, most of the network architectures propose uniqueness when it comes
to choosing their different protocols implemented by the different layers. Information travel-
ing on networks defined by these architectures permits different protocols on the different
layers, but on every hop, these protocols are limited to the ones chosen by the layer’s code.
For example, in the context of IP communications, we understand that a datagram is routed
on the basis of the information gathered from a certain internal router protocol, such as RIP
8  Opportunistic Networks

Figure 1.4  General network DTN architecture. The different levels in the DTN layer represent the possibility
of employing different routing protocols and different DTN behaviors. The different convergence
layers, such as Licklider Transmission Protocol (LTP) [9], provide communication among intermedi-
ate nodes when they meet in an opportunistic manner.

or OSPF, but we would not see it as feasible that the datagram could choose among these two
mechanisms in order to be routed.
Layered protocols do not only imply just a separation of responsibilities. Layers are placed
one after the other and define in which order things should be done. In TCP/IP for exam-
ple, routing algorithms are executed before the translation of logical addresses into physical
addresses while sending a data-gram.
We propose simple DTN architecture which is based on layers, but in a very particular way.
This architecture is fully compatible with the DTN proposals defined by the DTN architec-
ture [8] as will be described in the following sections. A layer-based model, according to the
definition explained in this section, implies delegation of tasks, encapsulation, and unique-
ness and should define a sequence of actions.
In Figure 1.4, our DTN architecture is depicted. An application layer represents the differ-
ent application which may employ the network. The DTN layer handles problems such as
addressing, routing, reliability and custody transfer, congestion, and security very similarly
to the DTN architecture with a subtle difference: the DTN layer may be different for every
application. Issues such as routing, reliability and custody transfer, congestion and flow con-
trol, and security may be treated differently depending on the application.

1.6 SYSTEM ARCHITECTURE ELEMENTS

In this section, the different system architecture elements are described.

RIT manager. Routing information is classified into ontologies, which represent the
different domains from the different applications which may coexist in a network. The
RIT manager is responsible for the access and storage of this information, which can
be created when nodes meet and exchange application information, or with the output
from active messages execution codes. The very same active messages may classify,
modify, access, or remove information from the routing information tree contacting the
RIT manager. Local nodes limit, remove, order, index, or purge the content of the RIT
depending on their resource constraints.
Local routing information generator. Local routing information may be created by the
local nodes for future use of routing algorithms. Algorithms like PRoPHET [10] need
 Mobile-Code-Based Opportunistic Networking  9

local routing information which should be updated at certain times and every time a
node is contacted. The information created by the local routing information generator,
is stored in the RIT following a certain and well-defined order in order to allow active
messages to find the necessary information for its routing decisions.
Network abstraction: A module called network abstraction is defined to permit local
nodes interact with different heterogeneous networks. Basic primitives, send, receive,
and neighbor discovery are available to the nodes. The primitives send and receive
allow transmitting and receiving bundles of information among nodes which can be of
three different types:
• Active messages: As introduced in Section 1.4, these are messages containing appli-
cation data. The content of these messages include application data, application
code, and routing code.
• Beacon messages: Beacons are sent at certain intervals of time to allow neighbors
to be discovered. These messages contain a node identification using Universal
Resource Identifiers (URIs) as defined in [11]. Along with the node identifier, basic
information that describes the node may also be included in the beacon messages.
Which information from the different fields in the RIT should be announced is not
a trivial issue. In order to be flexible enough, we allow the very same application
messages to flag the desired RIT fields among the RIT application branch to be
announced.
Scheduling and forwarding manager: Two queues for incoming and outgoing active mes-
sages are managed by the scheduling and forwarding manager. A prioritized scheduler
manages which active messages should be forwarded and accepted first and for dis-
carding purposes. For ordering or discarding active messages, we will also consider
information from other layers, like the application layer. Other proposals for ordering
DTN take into account only information from the DTN layer. We believe information
from the DTN layer is important, but information from the application layer is crucial
to distinguish which are important messages and which are not (Figure 1.5).

Figure 1.5  The elements that form the DTN proposed system architecture.
10  Opportunistic Networks

Pool of forwarding messages: Active messages are accepted by the local nodes in the pool
of forwarding messages following the delay and disruption paradigm store-carry-and-
forward. Storage reserved for these purposes is a key issue in opportunistic networks.
Routing algorithms must prevent from making epidemic decisions which may flood
buffers containing the pool of messages.
Routing and application code execution environment: As already explained, active
messages contain code for routing and application behavior purposes. In order to
execute this code, an execution environment must be included on every DTN node of
the network. In the following sections, the code execution environment is explained
in detail. Two examples of execution environments to be used in DTN scenarios are
presented.

In Figure 1.6, the architecture of our proposal is depicted. On one hand, we use mobile
nodes which define a delay tolerant network (DTN) by means of exploiting mobile node
opportunistic or scheduled contacts. Simultaneously, on the other hand, this network is
employed by active messages which are responsible for sensoring tasks. Every mobile node
carries a smart device which runs an execution environment for running active messages.
Data obtained from sensing tasks may travel along with the active message as part of its
data until the active message reaches a sink node. However, as shown in Figure 1.6, once
data are created, they can also become independent from the active message that created
it. Consequently, data delivery can be considered from the traditional point of view already
studied in the bibliography. It has one particularity, however: the communication paradigm
in this case follows a many-to-one or many-to-few model, where a small group of sink nodes
represent the destination nodes. In our scenario, the deployed mobile nodes act as DTN
routers, using algorithms which can be found locally in the mobile node and follow mobility
models, as explained in Section 1.12. Section 1.10 describes how these routing algorithms
can be updated by the very same active messages.
Figure 1.7 shows how the different actors involved in the proposal interact with each other.
These actors are:

Figure 1.6  A system architecture for DTN mobile code-based WSN.

 Mobile-Code-Based Opportunistic Networking  11

Figure 1.7  General scenario. Active messages travel from DTN node to DTN node creating data messages
which are sent to the sink nodes.

Mobile nodes: Robots in our proposal, these carry smart devices which control sensor
devices. They are capable of staying in the sensed zone for a given time and are able to
come back to the charging stations when batteries start to run out of power.
Active messages: These are software entities responsible for sensing tasks. They are capa-
ble of migrating from mobile node to mobile node in order to fulfill tasks requested by
users.
Execution environments: Active messages need execution environments for running,
which themselves run on smart devices. They are responsible for negotiating movement
model changes, as described in Section 1.12.
Code execution environment managers: These software entities remain on the code execu-
tion environments. They behave as intermediary actors between hardware sensors and
active messages. They are also responsible for negotiating movement model changes.
Data messages: Once sensing actions are performed, data are acquired. These data have as
a destination the sink nodes, responsible for the data processing. Data messages do not
travel inside active messages. They are independent pieces of information that travel
from mobile node to mobile node having the sink nodes as destination.
Sink nodes: These are destination nodes for WSN. These nodes are normally stationary
and are plugged, connected and located outside the sensed zone.

1.7 SECURITY CONSIDERATIONS

We are aware that security in opportunistic networks is a very important issue [12,13]. While
designing our architecture security has been fully taken into account, even if in this chapter it
is not directly covered, in this section we want to enumerate the list of security considerations
which should be studied. These mechanisms must guarantee the following properties:

Data confidentiality: The confidentiality of the data provided by the application sources
and sent to the destination nodes must be warrantied. Otherwise, an attacker could
perform some malevolent actions in order to eavesdrop on this information [14].
Data authenticity: Authenticity is vital to avoid any sending or injection of false infor-
mation destined for a DTN node. It is also important to remark that this authenticity
should not be linked to the identity of the node that generated the information [15].
Data integrity: The integrity is crucial to permit the detection of a state where a compro-
mised node has modified the data which come from the application sources, or the data
which it forwards to another node in a routing process [16].
12  Opportunistic Networks

Since the active messages are one of the pillars of our proposal, and since their code can
influence the movement and the behavior of a node, it is important to protect it to avoid
undesirable circumstances. In this case, we consider the following security properties as the
most appropriate.

Code confidentiality: The confidentiality of the code is a desirable property to avoid out-
side attacks, which can include eavesdropping on the radio channel, the goal of which
is to profile the kind of information that an application is gathering, and the procedure
used to obtain it. Although this requirement can be desirable in some cases, this is not
a mandatory property, and it depends on every particular application.
Code integrity: The authenticity of the code of an active message must be warrantied in
order to avoid an illegitimate active message code being executed in a node. Otherwise,
an insider or outside attack could inject a malicious active message with the purpose of
compromising the security of the infrastructure.
Node identification: Since the custody of a message is delegated to DTN nodes, the correct
identification of a node is a very important issue. A security mechanism should be
defined to avoid an illegitimate DTN node to impersonate a legitimate one.
Key management: In the case symmetric or asymmetric keys are employed in the security
mechanisms proposed in this section, a key management protocol should be defined.
Issues like revocations lists deployment are complicate tasks in intermittently connected
scenarios.

One of the most important aspects of our proposal is the use of the active messages as a way
of routing the code that must be used to take the measurements. This can be done by using
active messages and by sharing contextual routing information between the nodes. Thus, it
seems evident that security must be present in the routing procedures. We impose the follow-
ing minimum requirements to guarantee the security of the routing:

Routing data authenticity: The authenticity of the contextual routing data shared by nodes
must be preserved in order to avoid that any outside attacker injecting incorrect infor-
mation. This property can also be seen as a way to deter any potential outside attacker,
since they cannot provide this property. It is clear that this requirement does not allow
us to detect any malevolent data injected by an inside attacker.
Routing data integrity: Integrity must be preserved with the aim of avoiding an attacker
modifying the routing data, as this change is not detected by its receiver. Again, this
property cannot be preserved in the case of an inside attack, but it is also one way to
dissuade any potential outside attackers.
Active messages forwarding non-repudiability: The non-repudiability property in the pro-
cess of an active message forwarding must be warrantied from the source and the des-
tination perspective. This is usually known as strong fairness and allows us to detect
when a compromised node has completed a malicious action from the perspective of
the routing of the active messages. For this purpose, all the non-repudiability evidence
must be correlated to detect when an inside attack has performed a malicious behavior.
This correlation must be done in places considered points of trust.

1.8 REIFICATION

Using a network of mobile nodes as a generalization of robots for sensing purposes is not a
new idea [17]. Many proposals have been described or envisaged, hitherto making use of a
troop of mobile devices equipped with a bunch of sensors. Some of them even use the very
 Mobile-Code-Based Opportunistic Networking  13

same mobile node network to transmit the acquired samples to a data sink. Yet, most of these
systems are oriented to just one application, and all critical mechanisms, such as message
routing or the physical movement pattern, are fitted with the specific actions and main goals
of that particular application.
Being practical, and considering other network evolutions such as the Internet, it would
be much more useful to take advantage of such a physical infrastructure, which would allow
several applications to simultaneously use the mobile nodes and their sensors in their own
way. The benefits are clear: a general-purpose sensor mobile node network is easier when
reusable, and it allows new applications to be implemented after the mobile node network
has been deployed. Unfortunately, turning such systems into general-purpose, multiple
applications is not very easy. If several different applications coexist, which is the best
message routing algorithm for them all? If mobile nodes can alter their pathways to adapt to
some applications’ requirements, what if there are a handful of applications striving against
each other to make the node move under their command? These and other issues have to be
considered when designing, and can be very hard to solve.
In this chapter, we propose a general-purpose sensing mobile node network, addressing the
previously stated questions and many other issues surrounding the coexistence of applica-
tions and the routing of information. The rationale behind this proposal is to have mobile
code at two different levels. First is as the user application itself, which can move from mobile
device to mobile device as it suits the application best in order to accomplish its goals. Second
is at the message level, allowing the very same message to make the routing decision by itself,
and independently from other messages’ routing policies. The resulting system happens to be
a particular case of the wireless sensor network (WSN), i.e., a network of autonomous sen-
sors aimed at monitoring physical or environmental conditions that pass their data through
the network to certain locations or data sinks [18], working as a DTN (delay and disruption
tolerant networking) [19]. In a DTN, data can be sent between any two nodes of the net-
work, even when middle nodes cannot have concomitant communications. This is possible
by following an asynchronous store-carry-and-forward paradigm. A wide variety of data
can be acquired by means of sensors, ranging from temperature, humidity, or noise levels,
for example, to other higher-level data such as object or human recognition, movement pat-
tern detection, or plague detection; the DTN approach only broadens the scenarios of this
particular WSN, allowing a myriad of applications.
One could wonder why a DTN-based network would be better in this case than a more
traditional ad hoc or, while we are at it, MANET, network. The main reason for this is the
required density of nodes. A DTN approach, for example, does not need a figure of nodes
directly proportional to the sensing area to guarantee the operation of the system. Likewise,
the sensing area can be extended or shrunken depending solely on the running applications.
On the contrary, communication range and sensing area determine the minimum number of
nodes to be used in ad hoc networks. Once an ad hoc network has been deployed, it is very
difficult to extend the sensing area without adding new nodes.
The architecture and operation of this network is presented within the chapter, which also
provides details on how some issues have been resolved. Examples include how population
growth is controlled when cloning is considered, how dynamic multi-routing is achieved, or
how mobile messages can influence node movement respectfully and fairly to other messages.
Security has been considered as well, analyzing the threats and requirements, and giving
mechanisms to protect mobile nodes.
The results are significant, and the overall performance of the system has been found to
be very reasonable. A number of simulations have been done, from which some interesting
outcomes have been drawn, such as a significant improvement in terms of message delivery
ratio and latency. Furthermore, a successful proof of concept has been undertaken using real
14  Opportunistic Networks

physical mobile nodes to check the feasibility of the proposal. The network has good potential
for use in a variety of applications such as underwater environment sensing, unmanned aerial
vehicle networks, environment applications, disaster field on emergency scenarios such as
earthquakes recovery or terrorist attacks, mines seeking, urban search missions, community
development, machine surveillance, accurate agriculture, biological attack reconnaissance,
and many others.
The original contributions of this study are: a general-purpose multi-application mobile
node sensor network based on mobile code, a dynamic multi-routing schema for allowing
different routing algorithms for different applications and an application influenced
movement model for minimizing node stagnation and maximizing segregation.
There are interesting publications like [20], which use active messages to sense infor-
mation in mobile wireless sensor networks in the context of surveillance systems. Basic
primitives for active messages missions are restricted to combing instructions in specific
areas, while other useful ones, such as search missions, are not considered. Routing deci-
sions for active message forwarding are evaluated in terms of geographical information
solely, while other useful information such as congestion, aggregation, movement models,
or code persuasion are not taken into account. Data messages remain with the active mes-
sage until it reaches a sink point, so there is no distinction between how the data messages
travel and how the active messages do, which may be inefficient in some scenarios. As a
result, active message missions are tightly coupled to mobile nodes, and regions with high
concentrations of nodes are created. In order to avoid this, complex mechanisms must be
implemented.
Even though there has been considerable ongoing work on WSN and DTN, as it has just
been shown in this section, there is not yet a mobile sensor network flexible enough to fit
any scenario, including those lacking continuous connectivity, and allowing sharing of the
network with other applications. The next sections describe, discuss, and reify precisely a
system overcoming these limitations.
As seen in the previous section, there are a wide range of different proposals to implement
wireless sensor networks (WSN). Several issues must be taken into account to provide a
general-purpose, multi-application, mobile node-based WSN. These issues include sensor
placement, sensor code deployment, sensing actions, sensor updating, task scheduling,
routing, routing algorithms deployment, defining movement models, speed control, obstacle
avoidance algorithms, energy-saving strategies, data fusion/merging/overlap, data division,
scheduling and dropping, and finally, data processing.
In our proposal to implement a WSN, active messages play an important role in developing
these tasks. Active messages are messages which, besides data, also carry code. We describe
in this section the benefits of using active messages, which are executed on mobile nodes to
obtain a general-purpose wireless sensor network. In addition to an architectural description,
we explain interesting issues to be studied, as well as the limitations of our proposal.

1.9 PRIMITIVE SERVICES TYPES AND TASK DELEGATION

From the point of view of the user, we have identified three different potential uses for
employing the network. On one hand, the user would like to search for a given pattern in
a given zone in the studied area. On the other hand, the user would also like to perform a
sensor comb of a chosen area. Finally, the user would like to arrive at a given point in the
studied area and perform some type of unique concrete sensing. We distinguish these three
types of behaviors because it will affect the way active messages will be created, how they will
migrate, how they will be reproduced, and their communication.
 Mobile-Code-Based Opportunistic Networking  15

Searching active messages are source-cloned when created. This is to say that the number of
copies of the message is decided before entering the affected area. Source cloning minimizes
communication among mobile nodes and active messages. Other proposals noted include
active message migration at any given point of the studied area. However, population control
is a complex issue for solving, due to its high communication requirements. Therefore, since
mobile nodes suffer from limited communication and energy constraints, we see in source
cloning a good solution for the population control problem.
A second service includes the types of tasks, the aim of which is to comb a given area for
different sensing purposes. In this case, the active message must first approach the selected
area. Once it has arrived at the selected area, the active message can then clone itself into
other active messages. Control of the population is guaranteed, since clones cannot live out-
side the given area. This is what we call floating cloning, following an idea similar to the
context of network-based social applications found in [21] and [10]. One delicate issue is
that of the size of the area, as it should not be large, so as to prevent the appearance of highly
concentrated regions. This problem will be further discussed in Section 1.12.
Finally, we consider the situation in which an application wants to go to a specific place
and for concrete sensing. In this case, there is no need to do any source cloning or floating
cloning.
Applications, however, may want to define composite tasks containing different service
types. For example, an application may want to define a search behavior in which once the
active message finds its target, it can comb the surroundings for additional information. We
distinguish then, among behaviors. That is, which of the three different service types the
active message is following and also the tasks, the high-level instructions defined by the users.

1.10 DYNAMIC MULTI-ROUTING

In our proposal we have to differentiate among two types of routing. On one hand, active
messages jump from node to node, in order to physically arrive at different points of interest
to perform their tasks. This action can be considered a routing decision, in the sense that the
active message makes a decision among the potential neighbors, considering as well the pos-
sibility of staying in the current node.
We see this routing as part of the active message model, that is, a complex composite
movement model made up of the list of all the n different mobile nodes the active message
migrates to, as described in:

movement agent   mc0 , mc1 , , mcn 

On the other hand, sensor data retrieved by the active messages travels heading towards
the sink nodes by jumping from mobile to mobile node. In scenarios in which different appli-
cations coexist, a single routing algorithm may not be enough to handle all of the applica-
tions’ needs. This is why routing algorithms must take into account information from the
application layer, as well.
This leads us to the necessity of having different routing algorithms on the mobile nodes
for data routing purposes. Stored locally on the mobile nodes, the routing algorithms are
chosen depending on the application. Given an application message mess, and application
app, a local custodian node cj and its routing information tree (RITcj). This other routing can
be expressed as:

 
forwardToList  flocalrouting messappi , neighbourList  c j ,t  , RITcj , where
forwardToList  neighbourList  c j , t   c j
16  Opportunistic Networks

That is, routing decisions are made, taking into account the list of the current neighbors
and the context information.
There are many proposals which define very efficient solutions for the routing problem in
DTN networks for very different DTN scenarios. However, not all applications employing
the same WSN need their data to be routed in the same way. A single routing algorithm may
not work for various applications. The aim of our approach is not just to improve transmis-
sion time, but also to provide a flexible and generic DTN network capable of handling dif-
ferent routing behaviors. The routing algorithms are easily deployed on the different mobile
nodes by using the very same active message. Coexistence of different applications willing to
use different routing algorithms cannot be easily deployed with other classic DTN proposals
such as bundle protocol-based or Haggle-based ones.
An illustrative example of the need for different routing algorithms in a single DTN sce-
nario may be found in emergency rescue coordination applications. In disaster areas, dif-
ferent users such as police officers, firefighters, doctors, nurses, engineers, or rescue teams,
among others, along with portable devices such as mobile phones or tablets, may share
and use the interconnected network with a mobile robot wireless sensor network. Different
applications such as victim location, notification applications, fire coordination and pollu-
tion measurement, or radiation location may need different ways of information routing,
scheduling, dropping, or aggregating. For example, information that contains a notification
to a given user or mobile node may be optimally routed using a probabilistic routing algo-
rithm such as PRoPHET [10]. Alternately, routing decisions for information resulting from
sensor tasks may depend on the level of importance of the information seen from the point of
view of the application. If the information is important, an epidemic routing algorithm will
be used. Otherwise, it can be discarded if the information is obsoleted by some other present
in the same node.

1.11 AGGREGATION, SCHEDULING, AND DROPPING

Previously in this chapter, we commented on the advantages of dynamic multi-routing for

both levels of routing presented: active message migration and data routing. Traditional
routing protocols follow address-centric protocols, in which decisions are based on the
destination address. However, other proposals such as [22] also consider the fact that
data can be opportunistically aggregated and consolidated at the routing nodes, reduc-
ing the impact on the network. Data-centric routing protocols, that is, protocols which
make decisions in terms of possible future data aggregation or data consolidation, are
being considered in our proposal. In the same way, data routing could be influenced by
future nodes storage congestion and more concretely by potential scheduling or dropping
policies.
When facing a multi-application schema for a general-purpose wireless sensor network,
data aggregation is highly correlated with the different applications. It is quite complex to
provide a general solution for the aggregation problem for the different routing algorithms
of the various applications. The objective is to make routing algorithms aware of potential
future aggregation, dropping, or scheduling actions. Therefore, a combined approach of
application-based and context-aware routing algorithms is crucial for these purposes. We
are employing the dynamic multi routing application explained in the previous section to
include in the different routing protocols ways of obtaining information about aggregation,
scheduling actions, and dropping policies.
 Mobile-Code-Based Opportunistic Networking  17

1.12 APPLICATION INFLUENCED MOVEMENT MODEL

Mobile nodes on a WSN may follow different mobility models. There is a one-to-many rela-
tionship among a mobile node and an active message; that is, active messages are allowed to
congregate in a single mobile node. Other proposals like [20] suggest that both actors should
be tightly coupled in a one-to-one way. Applications which work in concrete areas could
create highly concentrated regions, which are difficult to recover. Our proposal contends
that such movement model should be as independent as possible from the rest of the mobile
nodes, in order to minimize energy consumption. From a global perspective, we intend to
make mobile nodes remain as widely spread out on the affected area as possible but favor-
ing the mobile node contacts. These constraints are affected by the active messages and even
by the application data which can subtly modify the mobile node movement model for task
accomplishment purposes.
This movement negotiation is handled by the code execution environment managers.
Active messages make movement requests to the code execution environment managers,
which in turn evaluate these requests in terms of the positions postulated. A queue in which
these requests are stored is defined locally in the node. To accept the movement change and
therefore introduce a new point in the mobile node queue, the point requested must be inside
a circle radius of kd, where d is the distance from the current position of the mobile node to
the previously scheduled waypoint. The bigger the factor k is, the more tolerant to mobility
the mobile node will be. The mobile node will retrieve the closest position value from this
queue until this queue is empty and then proceed with its movement model.
As an example, in Figure 1.8, a code execution environment manager receives a request
from three active messages to temporarily modify its movement model. On the lower part of
the figure, we see how the movement model is modified by the active messages.
In order to prevent a possible mobile node stagnation caused by queued points which remain
far from the mobile node, requests from the active messages can be rejected. In Figure 1.9, an

Figure 1.8  Movement model. Active messages request the local node to accept a movement change by sug-
gesting one or more points of interest.
18  Opportunistic Networks

Figure 1.9  Movement request protocol. An active message requests a point of interest, while the local DTN
node may or not accept this movement change.

active message performs a movement request, for example, Point of Interest 3 in Figure 1.8.
The code execution environment managers decide to accept it because it is inside the radius
kd. Once the mobile node has visited the other two points of interest, the queued Point of
Interest 3 remains outside the new circle radius kd. The code execution environment manager
informs the active message that its request is rejected and the latter migrates to another mobile
node, as seen in Figure 1.9.

1.13 ACTIVE MESSAGES: A DISTRIBUTED SENSING INFRASTRUCTURE

Our goal is to provide a general-purpose WSN, multi-application system. Different applica-

tions are represented by distributed streams of active messages performing different tasks in
the studied zone. In this section, we enumerate these different tasks and analyze the benefits
of using the active messages paradigm in the context of WSNs.
Active messages carry the code responsible for managing the sensor retrieval following
users’ requests. Tasks may be divided into sub-tasks, which can be delegated to other active
messages (clones) outside or inside the code execution environment, as seen in Section 1.10.
However, applications may at some point evolve in their needs. In a traditional WSN, this
evolution means sophisticated deployment actions in order to modify the way the informa-
tion is gathered, which are not always easy to achieve. These adjustments may be caused by
the applications need to retrieve information from additional sensors, discontinuing some
others not needed anymore by the application or even modifying sensor retrieval a posteriori
actions.
Active messages are an excellent option to solve these problems. Since the sensor code is
traveling with the very active message itself, it is guaranteed that the way the information is
gathered, plus a posteriori actions if needed, will be the ones desired by the user. Updates on
these codes are made automatically if requested by the application, while visiting the sink
nodes or while opportunistically contacting other active messages in the studied area.
The application itself knows best how its data should be routed. As seen in Section 1.11,
information retrieved from the mobile sensors travels from mobile node to mobile node in
data messages, following a behavior pattern set by the different routing algorithms installed
locally on every mobile node. When a new application is willing to use the sensor network,
it is possible to include an ad hoc routing protocol for the application. Active messages carry
new routing algorithms to the different mobile nodes they visit so that future bundles of data
may use them. This deployment of routing algorithms may be seen as additional tasks which
the active message must accomplish.
 Mobile-Code-Based Opportunistic Networking  19

Nodes when finding other mobile nodes exchange context information which will be used
by the different routing algorithms, for both active messages and data messages. Willing-to-
migrate active messages communicate with a code execution environment manager to query
for foreign context information in order to make routing decisions.

REFERENCES

[1] Aruna Balasubramanian, Brian Levine, and Arun Venkataramani. DTN routing as a resource
allocation problem. In Proceedings of the 2007 conference on Applications, technologies, archi-
tectures, and protocols for computer communications, pp. 373–384, 2007.
[2] K Scott, S Burleigh, et al. RFC 5050: bundle protocol specification. IRTF DTN Research Group,
2007.
[3] James Scott, Jon Crowcroft, Pan Hui, and Christophe Diot. Haggle: A networking architec-
ture designed around mobile users. In WONS 2006: Third Annual Conference on Wireless
On-demand Network Systems and Services (pp. 78–86). ARES Team. 2006.
[4] Dana Simonson, John Livdahl, and Rory A Smith. System and method for multicast communica-
tions using real time transport protocol (RTP), May 22, 2007. US Patent 7,221,660.
[5] Carlos Borrego, Gerard Garcia, and Sergi Robles. Softwarecast: A code-based delivery many-
cast scheme in heterogeneous and opportunistic ad hoc networks. Ad Hoc Networks, 55:72–86,
2017.
[6] Jiannong Cao, Xinyu Feng, Jian Lu, and Sajal K. Das. Mailbox-based scheme for mobile agent
communications. Computer, 35(9):54–60, 2002.
[7] D Waitzman. Rfc2549: Ip over avian carriers with quality of service, 1999.
[8] Vinton Cerf, Scott Burleigh, Adrian Hooke, Leigh Torgerson, Robert Durst, Keith Scott, Kevin
Fall, and Howard Weiss. RFC 4838, delay-tolerant networking architecture. IRTF DTN Research
Group, 2(4):6, 2007.
[9] M Ramadas, S Burleigh, and S Farrell. RFC 5326, licklider transmission protocol specification.
IRTF DTN Research Group, 2008.
[10] Carlos Borrego, Joan Borrell, and Sergi Robles. Hey, influencer! message delivery to social central
nodes in social opportunistic networks. Computer Communications, 137:81–91, 2019.
[11] Tim Berners-Lee, Roy Fielding, and Larry Masinter. RFC 3986, uniform resource identifier
(URI): Generic syntax, 2005. http://www.faqs.org/rfcs/rfc3986.html, 2005.
[12] Adrián Sánchez-Carmona, Sergi Robles, and Carlos Borrego. Improving podcast distribution on
Gwanda using Privhab: A multiagent secure georouting protocol. ADCAIJ: Advances in distrib-
uted computing and artificial intelligence journal (ISSN: 2255-2863). Salamanca, 4(1), 2015.
[13] Adrián Sánchez-Carmona, Sergi Robles, and Carlos Borrego. Privhab+: A secure geographic
routing protocol for DTN. Computer Communications, 78:56–73, 2016.
[14] Carlos Borrego, Marica Amadeo, Antonella Molinaro, and Rutvij H Jhaveri. Privacy-preserving
forwarding using homomorphic encryption for information-centric wireless ad hoc networks.
IEEE Communications Letters, 23(10):1708–1711, 2019.
[15] Adrián Sánchez-Carmona, Sergi Robles, and Carlos Borrego. Privhab: A privacy preserving
georouting protocol based on a multiagent system for podcast distribution on disconnected
areas. Ad Hoc Networks, 53:110–122, 2016.
[16] Naercio Magaia, Carlos Borrego, Paulo Pereira, and Miguel Correia. Privo: A privacy-preserving
opportunistic routing protocol for delay tolerant networks. In 2017 IFIP Networking Conference
(IFIP Networking) and Workshops, pp. 1–9. IEEE, 2017.
[17] Carlos Borrego, Sergio Castillo, and Sergi Robles. Striving for sensing: Taming your mobile code
to share a robot sensor network. Information Sciences, 277:338–357, 2014.
[18] Winston KG Seah, Zhi Ang Eu, and Hwee-Pink Tan. Wireless sensor networks powered by ambi-
ent energy harvesting (WSN-HEAP)-survey and challenges. In 2009 1st International Conference
on Wireless Communication, Vehicular Technology, Information Theory and Aerospace &
Electronic Systems Technology, pp. 1–5. IEEE, 2009.
20  Opportunistic Networks

[19] Russell J Clark, Evan Zasoski, Jon Olson, Mostafa Ammar, and Ellen Zegura. D-book: a mobile
social networking application for delay tolerant networks. In Proceedings of the third ACM
workshop on Challenged networks, pp. 113–116, 2008.
[20] Edison Pignaton De Freitas, Tales Heimfarth, Ivayr Farah Netto, Carlos Eduardo Lino, Carlos
Eduardo Pereira, Armando Morado Ferreira, Flávio Rech Wagner, and Tony Larsson. UAV
relay network to support WSN connectivity. In International Congress on Ultra Modern
Telecommunications and Control Systems, pp. 309–314. IEEE, 2010.
[21] Jörg Ott, Esa Hyytiä, Pasi Lassila, Tobias Vaegs, and Jussi Kangasharju. Floating content:
Information sharing in urban areas. In 2011 IEEE International Conference on Pervasive
Computing and Communications (PerCom), pp. 136–146. IEEE, 2011.
[22] Ruhai Wang, Zhiguo Wei, Qinyu Zhang, and Jia Hou. LTP aggregation of DTN bundles in space
communications. IEEE Transactions on Aerospace and Electronic Systems, 49(3):1677–1691,
2013.
Chapter 2

Opportunistic Emergency Scenarios

An Opportunistic Distributed Computing Approach
Carlos Borrego Iglesias
Universitat de Barcelona, Barcelona, Spain

CONTENTS

2.1 An Opportunistic Approach for Mobile-Code-Based Distributed Computing ...........21

2.1.1 Integrating DTN WSNs in Grid Computer Infrastructures Using
Mobile Code ................................................................................................22
2.1.2 Grid Job Management................................................................................. 24
2.1.3 Store-Process-Carry-and-Forward Paradigm................................................ 24
2.1.4 Processing Models....................................................................................... 25
2.1.5 Storage........................................................................................................ 26
2.1.6 The Routing Issue........................................................................................ 27
2.1.7 Implementation........................................................................................... 27
2.2 Opportunistic Emergency Scenarios Applications...................................................... 28
2.2.1 Scenario Description.................................................................................... 29
2.2.2 Dynamic Routing and Routing Algorithm Deployment............................... 30
2.2.3 Alleviate DTN Congestion........................................................................... 32
2.2.4 DTN Lifetime Control................................................................................. 32
2.2.5 Dynamic Prioritized Scheduling ...................................................................33
Note���������������������������������������������������������������������������������������������������������������������������������� 34
References.......................................................................................................................... 34

2.1 AN OPPORTUNISTIC APPROACH FOR MOBILE-CODE-BASED

DISTRIBUTED COMPUTING

Distributed computing in general and grid computing in particular [1, 2] has consolidated
as a technology capable of solving some of the most challenging scientific projects of our
century. The needs of these projects usually include complex computation of data obtained
from different sources and stored in large storage resources. The main goal of grid comput-
ing, precisely, is to share these resources among different institutes and virtual organizations
across high-speed networks and to distribute and coordinate its processing [3].
Wireless sensor networks (WSNs), on the other hand, are a technology that can be very use-
ful when it comes to acquiring and transporting data collected in widely spaced areas. These
networks consist of different nodes carrying different sensors along with autonomous computa-
tional devices which transmit data through the network to some specific locations or data sinks.
This chapter analyzes how both technologies, grid computing and wireless sensor net-
works, can be combined into an integrated WSN and computer grid infrastructure allowing
new functionalities. The cornerstone of this conjugation is using delay and disruption toler-
ant networking (DTN) concepts [4, 5] along with mobile code to create an intelligent grid

21
22  Opportunistic Networks

network [6, 7] capable of routing and managing processes depending on the context. Some
other recent proposals integrate WSNs and grid computing as well. However, our proposal
comes from the network perspective. We consider WSNs nodes as intermittent connected
nodes, containing asymmetric bandwidths, long and variable latencies, and ambiguous mobil-
ity patterns. This new perspective contributes to the creation of a novel concept of intelligent
grid computing networks, going beyond the possibilities of the reviewed literature in some
current scenarios, and providing promising prospects for supporting future grid services.
The routing decision [8] making and execution policies travel with the messages, instead
of being static and exactly the same for all nodes. These policies, in the shape of mobile code
[9], can take into account the context of the nodes to choose the behavior that fits best in
each situation [10]. All in all, the system acts more like an ant colony, with differentiated
autonomous parts acting locally but with a cooperative aim, rather than a traditional and
more inflexible system. Thus, using mobile code makes the grid network an intelligent sys-
tem, pliable enough to adapt to new scenarios of grid computing. However, the proposed
system cannot be considered a silver bullet for all grid computing. In highly connected grids,
with low latencies and where data do not need to be processed before getting to the execu-
tion destination, mobile code would introduce an unnecessary extra overhead and other
unwanted side effects.

2.1.1 Integrating DTN WSNs in Grid Computer Infrastructures Using

Mobile Code
In this section we propose a way to integrate intermittently connected WSNs in computer
grid infrastructures. On one hand, the data acquired by the WSNs can be processed and
stored using traditional grid services inside connected regions. On the other hand, and as a
more challenging option, sensors can be considered as grid sources of data commanded by
the very grid users themselves. We propose the job behavior that evolves beyond the tradi-
tional read-process-and-storage model – that is, reading from a data source, processing, and
storing the result – to a possibility of having a create-read-process-and-storage model, in
which jobs are able to create their input information.
For both ways of integration, there is a challenging issue which must be solved. Data must
arrive from the WSN to nodes connected to the grid, known as sink nodes. However, con-
nectivity among the nodes belonging to a WSN can rarely be available, due to large latencies,
high error rates, or even very small bandwidths. Examples of these scenarios include outer
space sensors and computer facilities, and isolated environment sensors.
Traditional grid scenarios use different types of data sources which include large hadron
colliders like the LHC [11], biomedical equipment or supplies for studies like [12], or even
simulated data obtained from the very same grid resources in the context of the same domains.
Most of them are constantly connected to the Internet or have a straightforward way to store
its data to grid connected resources. Likewise, grid services are implemented based on networks
with a constant Internet connection, taking for granted a continuous network connection.
By considering WSNs as a potential grid source of information, we analyze in this study
how to include WSNs in grid computing, taking into account that they may be not fully per-
manently connected to the rest of the grid infrastructure. The result is to obtain what we call
a delay and disruption tolerant computer grid source.
In Figure 2.1, we can see how grid infrastructures and WSN technologies may coexist.
On the left-hand side of the figure, a WSN is depicted connected to a grid infrastructure by
a computing element, a door to the sensor network. This sensor network is then seen as a
grid source becoming grid-wide available. The WSN is composed by several sensors which,
as a global network, are intermittently connected to the Internet and could have a poor
 Opportunistic Emergency Scenarios  23

Figure 2.1  Coexistence of WSN and grid infrastructure. On the left-hand side, an intermittently connected
sensor grid is connected to a permanent connected grid infrastructure.

connection among them. Some of these sensors may adopt ambiguous mobility patterns and
therefore act as DTN data mules [5], sensors in movement which gather information from
other sensors which act as opportunistic router nodes.
The traditional DTN scheme follows the well-known store-carry-and-forward paradigm
[5]. We introduce a new paradigm: the store-process-carry-and-forward. We can benefit from
the fact that in DTN the information can be blocked waiting in some DTN node which tem-
porarily is disconnected. This waiting time can be extremely long, and we can make good use
of it. If the application data contain information to be processed, we could make the local
node process the data while it waits for some other node to arrive.
Distributed computing application information – that is, information created on behalf of
a user and retrieved from the sensors – is carried inside messages from the sensors back to
the computing element. Custodian nodes – that is, nodes which custody application informa-
tion until it can be forwarded to another node – handle delays and disruptions. The applica-
tion layer will create application data which will be included in one DTN message. These
messages will stay in a custodian node until they are able to migrate to another one. Since
delays in DTN networks can be huge, messages containing application data may wait for
a long time. Once a potential custodian node becomes available, the message will try to be
forwarded to the newcomer custodian node.
As seen in Figure 2.2, traditional grid layers – application layer, collective layer, resource
layer, connectivity layer, and fabric layer – can be completed by adding an execution

Figure 2.2  Grid layers.

24  Opportunistic Networks

environment layer. Mobile code can be seen as an alternative to the grid classic fabric layer.
Its aim is to provide an interface to local resources, the sensors. There is a subtle difference
with traditional grid architectures. This alternative to the fabric layer in this case is directly
connected to the application layer in order to allow applications direct access to the informa-
tion retrieved by the local sensors present in the nodes.

2.1.2 Grid Job Management

Information created in grid sources are processed using a computing infrastructure. Jobs
input data in experiments like ATLAS [13] can reach to be tens of thousands of times larger
than the output. In scenarios in which we include the grid sources, such as delay and disrup-
tion WSNs, we believe that computing locally may save loads of data transfer. This is espe-
cially interesting, since these transfers are very expensive.
In the case in which WSNs are isolated, for example a sensor network in deep space,
bringing data outside the network in order to compute it can be really expensive. Instead, we
propose a model in which raw data, information extracted directly from the grid sources, are
kept and processed locally.
Traditional grid users define their jobs by specifying their executables, their input data,
their output data definition, requirements such as the site the user wants to send the job to,
and the virtual organization. The computing element receives the job specification along with
the files the user has decided to send to the site. This information is passed to the different job
managers. A typical job manager would receive a job request and would submit the job to a
local queuing system, such as pbs [14].
In our approach for sensor grid scenarios, the job manager’s behavior would be differ-
ent. We have defined and implemented a computer element job manager which, instead of
submitting a job to a queuing system, will create an active message and will launch it to the
WSN. Active messages are messages capable of performing processing on their own and
consequently behave autonomously. These messages will travel from node to node in order
to accomplish their tasks.
Creating this active message is quite a challenging issue. We need to clearly specify its
behavior in terms of routing decisions, sensor retrieval, data management, and task schedul-
ing. The user, by means of the job description, can explicitly specify these issues; otherwise,
the job manager will be responsible for properly creating it.
Looking more closely at the intelligence of this message, from one hand we have to dif-
ferentiate between the application code, that is, the behavior the active message will require
which defines the actions and tasks it will perform. These tasks include, for example, search-
ing and combing actions in the wireless sensor network area and processing data [15]. As
a result, application data are created. From the other hand, the active message will need
to make routing decisions whenever finding different nodes on the networks. Traditionally,
DTN approaches leave routing decisions for DTN nodes. In our proposal, intelligence comes
from allowing routing code to travel with the very active message, allowing different applica-
tions employing the same WSN to use different routing algorithms. Finally, both input and
output may need to travel with the active message. This mapping is represented in Figure 2.3.

2.1.3 Store-Process-Carry-and-Forward Paradigm
The active message implements the store-process-carry-and-forward paradigm:

• Once an active message arrives to a node, it waits stored in a queue of messages, until
another node is available for forwarding.
 Opportunistic Emergency Scenarios  25

Figure 2.3 Active message creation. The grid job manager creates an active message from the grid job
specification.

• When the active message is waiting for a node to be forwarded, if the input files of the
grid job are available, process of data can be started.
• When the message is forwarded, it behaves like a data mule [5], making an active carry,
i.e. carrying information containing the grid job output. Instead, a custodian node in
motion, as it could be a robot, satellite or an animal, performs a passive carry when
moving. In this case, no software entity is performing the transport itself. Instead, there
is a physical movement of the custodian node who produces the carry action.
• The very active message contains internally a routing algorithm to decide when and
where to be forwarded. The routing algorithm is part of the active message. This way
we allow different applications using the same infrastructure to behave differently
when it comes to routing decisions, contrary to classical networks such as TCP/IP, in
which routing static elements implement and perform the routing decisions.

2.1.4 Processing Models
Bringing back the concept of data processing while waiting to migrate, it could happen that
when the message is able to be forwarded from the current custodian node, the code execu-
tion has not yet finished.
We have three behavior possibilities here to define how to handle job management:

• Run & Go. When a new node is available for forwarding, jobs do not abort their
execution. Active messages try to be forwarded once the job has finished; however, the
new node still might not be available for forwarding. This option is used when node
density is high. In Figure 2.4, a new node appearance is represented by an arrow. In
Figure 2.4(a) concretely, a job finishes its execution and waits for the next node to be
forwarded.
• Abort & Go. Jobs could be aborted when a new custodian node appears, losing all
the processed information. The active message could eventually discard the computing
effort already done, finding it more convenient to be forwarded to a new node. This is
represented in Figure 2.4(b). Jobs will start again in the new custodian from scratch.
This option is useful when contacts are scarce and processing cost is low.
26  Opportunistic Networks

(a) (b) (c)

Figure 2.4  Processing models.

• Stop, Go, & Resume. Jobs stop their execution when they are aware of a new suitable
custodian available, forwarding themselves to the new custodian and resuming their
processing in the new node. As seen in Figure 2.4(c), the job is stopped but resumed in
the new node.
• Weak mobility is limited to the code and the data state. When arriving in the new
execution environments, the code would be restarted.
• Strong mobility allows code to be moved at a very low level, such as stack pointers
and instruction pointers to different execution environments.

Before the active message code is invoked, its job behavior is queried. This behavior defines
how an active message wants to proceed if the message is ready to be forwarded but the pro-
cessing is not yet finished. The EEM is responsible for finding the application code present in
the active message and for executing it. This code represents the way the active message will
behave in terms of sensor retrieval and task accomplishment in a custodian node.
In this way, the EEM has control of all the computing running on the local execution envi-
ronment and can stop, suspend, or prioritize any computation from any active message. Jobs
prioritization is not covered in this chapter, but it is a very interesting open issue.

2.1.5 Storage
Storage service in grid computing is responsible for managing and holding data generated by
grid sources or computing jobs as well as to provide this data grid-wide upon request. In our
proposal, we have two options for creating the active message concerning the storage of the
information created and/or processed.

• Store the information in the disconnected region. This information will be physically
stored in one of the DTN nodes from the disconnected region. The very active message
will choose which would be the physical location of the data, which could be any of the
DTN nodes the active message will visit, a subgroup of them, or a given one.
• Store the raw data outside the disconnected region. Instead of leaving the data inside
the disconnected region, the data is routed to a DTN gateway, a special node that
belongs to both the disconnected region and the connected region. Subsequently, the
data is stored in a storage element outside the disconnected region.
 Opportunistic Emergency Scenarios  27

By any means, information needs to be forwarded from one node to another, traveling all
around the disconnected region. Thus, the storage problem becomes a routing problem,
hence, finding the appropriate DTN node to leave the data once created. Many articles such
as [16] cover the still-open topic of DTN routing.
In any case, the file, after being stored, needs to be cataloged so it can be located in the
future. The active message will need to reach the connected region in order to inform the file
catalog about the information placement.

2.1.6 The Routing Issue

In our intelligent system, routing is the decision process in which the best path inside a given
network is calculated. In traditional networks, routing decisions aim to minimize latency
of message delivery and improve the delivery ratio, without overlooking local and network
resource consumption. In the proposed scenario, routing is a very delicate issue. Once an
active message arrives at the WSN, latency time to arrive at its destination and the delivery
ratio are important.
However, maximizing task performance in our scenario could be more important than
improving latency times or delivery ratio. This performance is different from application to
application and depends on several factors such as processing time, network congestion, flow
control, etc. Letting the very same application perform the routing decision can improve task
performance.
As an example, in Figure 2.5 an active message performs a combing task for zone 1. This
message must choose among two potential custodians, S2 and S3, in order to be forwarded.
The routing code for this active message chooses custodian S2 to improve zone 1 combing
task over choosing S3, even if this latter seems to have a direct contact with its destina-
tion. The very active message routing code prioritizes task completion over delivery latency.
Without an application perspective, this decision would be very difficult to make by a local
routing algorithm running in a custodian node. This allows the applications to intelligently
employ the network in a very personal way.

2.1.7 Implementation
The execution environments on which active messages run are called platforms. We need to
implement on every sensor grid node an execution environment platform to let active mes-
sages carry grid-level information to run on. In order to implement active messages described
in the previous sections, we need some special platforms: besides being capable of executing

Figure 2.5  Dynamic routing. An active message selects node S2 as its next custodian node.
28  Opportunistic Networks

code, platforms must allow the code to be forwarded from one sensor node to another, stop-
ping their code and resuming its execution upon arrival, and permit the ability of carrying
their data and state with them.
Among the different available execution environment platforms, we decided to use
Mobile-C [17], a Foundation for Intelligent Physical Agents (FIPA) [18] complaint platform
for mobile C/C++ codes in networked intelligent mechatronic and embedded systems. The
benefits in comparison to other execution environments like Jade based on Java are mainly
achieved through the reduction in memory consumption. Moreover, in general terms, the
impact on system resources is very small.
We have implemented a job manager1 in the context of gLite grid middleware [19]. This
job manager accepts traditional grid job requests and, by parsing jobs requirements, accord-
ingly creates active messages. We are using gLite’s grid job description language to allow the
user – besides the traditional InputSandbox and OutputSandbox – to specify the application
code and the routing code in the creation of the active message. One example is described as
follows:

JobType = "ActiveMessageJob";
ApplicationCode = "combing.cc";
RoutingCode = "myprophet.cc";
StdOutput = "output.txt";
StdError = "error.txt";
InputSandbox = {"combing.cc","myprophet.cc"};
OutputSandbox = {"output.txt","error.txt" };

Once active messages are created, they travel along with the grid’s input data if any, from
node to node, performing their tasks and retrieving information data. Execution environ-
ments custody active messages until another node becomes available for forwarding. The
very same active message chooses its node path by executing the Mobile-C MC – AddAgent()
method. Afterwards, it waits until a new node becomes available. Considering DTN scenar-
ios, the active message can stay in this state for a long time. We want to let active messages
choose for themselves their next hop, that is, the routing decision. This is done by modifying
the Mobile-C code to allow for having a well-known method for these purposes.

2.2 OPPORTUNISTIC EMERGENCY SCENARIOS APPLICATIONS

The widespread use of portable devices that are generally equipped with wireless-enabled
communications, GPS receivers, and/or touch screens has remarkably improved outdoor
applications in a great variety of scenarios. Although the most common network configura-
tion is ad hoc, or mobile ad hoc (MANET), new communication paradigms are emerging
to fill the void for some specific settings. This is the case of Delay and Disruption Tolerant
Networking (DTN) [3], which is extremely useful when no concomitant network links con-
nect the source and destination at transmission time. Emergency and disaster recovery sys-
tems are an example of an application domain where having no network infrastructure
makes DTN significantly extend possible applications. Applications based on DTN can
coexist with other solutions in order to create network infrastructure and restore network
connectivity. Furthermore, the DTN approach provides cheaper, easier, and ready-to-use
deployment solutions.
Although DTN has strong foundations such as the Bundle Protocol [20], and many groups
have been working on its formalities for some years, there are still a number of issues to
solve. Some of the most problematic issues include routing, lifetime control, and security,
 Opportunistic Emergency Scenarios  29

which need solutions quite different from the ones normally used on the Internet. The ratio-
nale for this is that the diversity of applications running on such a limited connected network
calls for a number of different mechanisms to solve the specific problems presented by each
application. As opposed to what happens on the Internet, no general-purpose mechanisms
exist which satisfy the requirements of all applications at once. A possible way of facing this
challenge is by adding code to every bundle of information sent over the network to make
(autonomously and in a context-aware fashion) all the decisions regarding the transportation
of that particular piece of data. Mobile code [21] is a well-known technology designed for
precisely this.
Mobile code is a good candidate for implementing a data-driven approach to DTN net-
works. This technology is able to migrate with their code and data from host to host and
to continue their execution upon reaching its destination. Mobile code needs execution
environments to run the execution environments. Mobile-C [17], Jade [18], and Agentscape
[22] are popular examples of platforms supporting mobile code. However, these platforms
cannot be directly used for implementing DTN networks. To begin with, forwarding proce-
dures have to be redesigned for the decision-making on which next intermediate node, also
known as the custodian node, to be forwarded to. We propose leaving this decision to the
bundle itself.
We present in this chapter an extension to the Bundle Protocol based on mobile code
to implement a heterogeneous data-driven DTN. While fully respecting the Delay-Tolerant
Networking Architecture [23], the keystone of this integration of technologies is to move the
routing algorithm from the host to the bundle. We will provide details on how this approach
significantly improves these networks. We propose a scheduling infrastructure which mini-
mizes delays as well as improves reliability by prioritizing bundles carrying important appli-
cation data. We also describe several mechanisms to solve specific issues with DTN networks
concerning routing, congestion, data aggregation, routing code deployment, and DTN life-
time control.
Emergency and disaster recovery scenarios are very convenient to demonstrate the new
possibilities of this combination of DTN concepts and mobile code. This study uses the case
of the Mobile Agent Electronic Triage Tag (MAETT) [15], to illustrate how applications can
profit from this new and improved architecture. However, the results can be used in many
other domains.

2.2.1 Scenario Description
Disaster recoveries after emergencies, such as terrorist attacks or meteorological calamities,
are difficult to conduct. Connected areas may become precipitously disconnected. Using DTN
networks is an excellent way of rapidly deploying communication networks. Other studies,
like [24], already use DTN networks to coordinate victims from emergency scenarios. In this
section we will define a scenario based on [24] to show the advantages of using code block
extensions in DTN networks.
Different users such as police, firefighters, doctors, nurses, engineers, and rescue teams,
among others, along with their portable devices such as mobile phones or tablets, create the
intermittently connected network. Opportunistic contacts among the different users permit
the different applications to utilize the network for very different purposes.
Rescue applications conducted by doctors, nurses, and rescue teams classify the different
victims during an initial evaluation of their health condition while in the confines of the
emergency area. Statuses attributed to victims are 0 for deceased, 1 for seriously injured,
2 for injured, and 3 for mildly injured. This process is called triage. Figure 2.6 depicts this
application.
30  Opportunistic Networks

Figure 2.6 Different users in an emergency scenario create a DTN network. A sink node, the Emergency
Coordination Center, collects the information from the different victims. A critical area where high
density of victims is located on the top-right part of the figure.

To manage and organize injury statuses and locations, we propose sending bundles with
the information of every victim that has been classified. The goal of this application is to
allow this information to arrive as soon as possible at the Emergency Coordination Center,
hopping from device to device when a shorter route is detected.
Other applications such as notification applications, wireless sensor applications, applica-
tions used for firefighting, pollution measurement, or radiation detection could employ the
same network. The coexistence of these applications is depicted in Figure 2.6. Coexistence
of applications by allowing different users to share a single network decreases cost and
creates a favorable atmosphere for node contacting. However, this coexistence of applica-
tions and having simultaneous users introduces some challenges that will be described in
this section.

2.2.2 Dynamic Routing and Routing Algorithm Deployment

In emergency scenarios, different applications may coexist. For example, data from victim
rescue application operations may share the network with wireless sensor data from firefight-
ers or other emergency response teams. Different applications may need different types of
routing algorithms. For example, information that contains a notification to a given user or
mobile node may be most favorably routed using a probabilistic routing algorithm such as
PRoPHET [16].
Alternately, routing decisions for information resulting from sensor tasks may depend on
the level of importance of the information seen from the point of view of the application.
If the information is important, an epidemic routing algorithm will be used to improve the
delivery performance. Instead, it can be discarded if the information is made obsolete by
other information present in the same node. The routing decision may travel with the bundles
themselves instead of being static and exactly the same for all nodes in the network. These
 Opportunistic Emergency Scenarios  31

policies, in the shape of mobile code, may consider the local context to choose the behavior
that fits best in each situation. Routing algorithms are carried using the bundle routing code
blocks described in the previous section.
We may differentiate among three types of routing algorithm deployment in emergency
scenarios.
First, the bundle routing code (BRC) is the routing algorithm employed by the bundles
to choose where the bundle should be forwarded to. These routing algorithms may travel
along with the bundles themselves, as explained in this section. However, we propose an
alternative way of deploying this code in case this paradigm cannot perform properly for
a given application or a given DTN scenario. New applications arriving to the emergency
zone may perform a routing algorithm deployment before sending its data. After the deploy-
ment, bundles may carry Routing Code Blocks with references to already deployed routing
algorithms, setting the Reference/Value bit and expressing the path to the routing algorithm.
During the deployment phase a broadcast [25] bundle containing a Routing Code Block with
the routing algorithm is sent to the network. Additionally, the “Local copy” bit should be set
to allow the routing algorithm to be copied locally to the Routing Information Tree (RIT) in
the appropriate path indicated in the RIT path field.
Second, some routing algorithms need to periodically execute code on every custodian
node. For example, a notification application for users inside the emergency area may
employ a probabilistic routing algorithm such as PRoPHET [16] to route its information.
This kind of routing algorithm needs to execute a code on the custodian nodes to update
delivery probabilities for every node contacted as well as to decrease these probabilities if
the nodes are not contacted. These codes may be deployed by including a Custodian Routing
block in a bundle. In this bundle, the periodicity of the code should be indicated by using
the Periodicity field.
Third, an application may install in the custodian nodes an aggregation function for
network traffic reduction and energy consumption saving purposes using the Custodian
Routing block. Data aggregation is an essential paradigm for DTN emergency scenarios
such as DTN wireless sensor networks that sense information from the emergency area.
The aim of this procedure is to merge the data coming from different custodian nodes,
eliminating possible redundancy, improving DTN congestion by minimizing the number of
transmissions and therefore improving energy consumption. For example, different data
belonging to an application from a contamination sensor network placed all along the
affected area may be aggregated following different mathematical functions such as aver-
age, summation, maximum, etc. These strategies are fully application dependent; there-
fore, a general solution for every application employing the network will probably be
inefficient.
Application routing algorithms may recognize these advantages in advance and take them
into account as an extra routing criterion. Address-centric routing algorithm approaches –
that is, finding the shortest path between the source and the destination can be evolved or
combined with an information-centric approach in which data is consolidated by applica-
tion-dependent redundant functions.
This paradigm shifts the focus from the traditional address-centric approaches of network-
ing (finding short routes between pairs of addressable end nodes) to a more information-
centric approach (finding routes from multiple sources to a single destination that allows
in-network consolidation of redundant data).
For these purposes, bundles may choose to delay its routing for a period of time in order
to maximize application aggregation. Since data aggregation cannot be performed among
different applications, this aggregation delay is an application-related issue and may be cal-
culated in the bundle routing code.
32  Opportunistic Networks

2.2.3 Alleviate DTN Congestion

In scenarios like disaster management, network congestion avoidance is a top priority. We
consider the routing problem to be an optimal resource allocation challenge. We propose
using the RIT update code block to allow the bundles to inform local and outer custodian
nodes about different application-related issues.
Each time a user leaves the Emergency Coordination Center, they set a timer indicating
when they expect to return. This timer, called Time to Return (TTR) by the authors of [15],
automatically decreases its value as time goes by. If a user detects another user with a smaller
TTR, it will try to forward its information to this user and thus improve the expected time
of arrival at the Emergency Coordination Center. For example, in Figure 2.7, we see how the
sender application from the first custodian node chooses the lower custodian node (doctor0)
as its next-hop because it has been informed by another bundle (Bundle2) about a low Time
to Return to the Emergency Coordination Center. Forwarding bundles to the upper custo-
dian node (doctor2) could cause congestion.
This is carried out by allowing the bundle agent to execute a concrete method present in
the code included in the bundle extension, (the RIT update code extension).
This information will be used by other bundles to update the information on outside
variables which may cause congestion, which could then be employed by their routing algo-
rithms. Congestion is highly correlated with routing algorithms, and the infrastructure pro-
posed is an excellent way of informing the different routing algorithms from outer platforms.

2.2.4 DTN Lifetime Control

In emergency scenarios, lifetime control may vary from application to application. We pro-
pose a mechanism beyond the Bundle Lifetime field to indicate when the bundle’s payload
is no longer valid. This mechanism is code-based and allows the application to control the
bundle’s lifetime from an application perspective. Using the lifetime code extension in con-
junction with the RIT update code extension, applications are allowed to cancel application
flows and inform both the intermediate custodian nodes and the sender application.
In Figure 2.8 we see how an application sends a bundle containing some application data.
Once this bundle has arrived at the fourth custodian node, it executes a special code using

Figure 2.7 Bundle2 forwarded from doctor0 node carries information about doctor0’s Time to Return (TTR).
After reading this information, Bundle1 message on top chooses doctor0 as its next custodian.
 Opportunistic Emergency Scenarios  33

Figure 2.8  Node node1 executes the lifetime code extension for Bundle1, deletes it, and sends an information
bundle with a RIT update code extension to cancel the original bundle flow in the intermediate cus-
todian nodes and the sender custodian node.

the lifetime code extension to recognize that the application information carried is no longer
valid. This could be due, for example, to some context change which could be evaluated
only locally and from the point of view of the application. In this case, the bundle generates
a control message using the RIT update code extension to the sender application, and while
visiting every DTN node it will inform all of the bundles containing application information
from the same flow (if any). When the bundle arrives at the source DTN node, it will defi-
nitely cancel the application flow. Since local context information differs from one custodian
node to another, we propose a solution flexible enough to allow the middle DTN custodian
nodes to cancel an application flow, improving DTN congestion as bundles are cancelled.

2.2.5 Dynamic Prioritized Scheduling

Since delays in DTN networks can be very big, bundles containing application data may
pile up at the DTN layer. Once a potential custodian DTN node becomes available to a
given custodian node, the latter will try to forward some of their bundles to the contacted
custodian node. It is possible that other bundles are intending to be forwarded to the same
destination custodian node at the same time. Policies like FIFO may not be enough for some
DTN scenarios. To regulate these situations, a prioritized scheduler inside the custodian node
is defined. This scheduler will decide bundle order forwarding and discarding policies.
In the scenarios described, while two users inside the emergency area are next to each other,
bundles containing information about victims, for example, may be forwarded to another
custodian node. Usually, the time it takes to forward this information is minimal due to the
users being in motion. This is why it is useful to first forward bundles containing victims with
a given state over others.
In the same way, if the buffer intended for storing bundles gets full, there must be a crite-
rion for choosing which bundles should be discarded. The majority of the proposals are static
in the sense that all custodian nodes behave in the same way. Our proposal provides criteria
for dropping bundles in the case that the buffer gets full, but with a valuable dissimilarity.
We let bundles themselves carry these criteria and update custodian nodes in order to permit
application dynamism when it comes to bundle prioritizing or dropping.
Our proposal consists of including bundle agents running on user’s devices, a prioritized
scheduler that orders bundles while taking into account the victim’s state. The prioritized
scheduler must then differentiate among the varying victim states.
It is necessary to consider that the criteria used to schedule bundles could change at any
given time. Let us consider the possibility of a bigger disaster which could eventually collapse
disaster recovery services. In this case, disaster recovery coordinators could decide that the
34  Opportunistic Networks

information on the casualties, that is, victims of level 0, are no longer important. Or, in a
more pessimistic scenario, the coordinator could be forced to let victims of level 1 die. These
criteria could change dynamically within the time of the disaster.
By using the bundle priority extension block, described in Section 2.2.5, bundles carry
information not only about the victims but also about the criteria to schedule themselves. If
the bundle agent finds that the criteria in a platform has been made obsolete by the one car-
ried by the bundle, it will update itself in the RIT.
Bundles belonging to different applications should not compete directly among themselves.
A round robin structure containing different queues for every application is proposed. The
platform will go around this structure obtaining the first n elements of each queue, depending
on the weight assigned to every application. This weight is determined by how important the
application is and how it is controlled by the local bundle agent.

NOTE

1 Source code can be found at http://ccd.uab.es/_cborrego/active-sensor-grid-src-1.2.tar.gz

REFERENCES

[1] Carlos Borrego and Sergi Robles. Relative information in grid information service and grid
monitoring using mobile agents. In 7th International Conference on Practical Applications of
Agents and Multi-Agent Systems (PAAMS 2009), pp. 150–158. Springer, 2009.
[2] Carlos Borrego and Sergi Robles. A store-carry-process-and-forward paradigm for intelligent
sensor grids. Information Sciences, 222:113–125, 2013.
[3] Carlos Borrego, Sergi Robles, Angela Fabregues, and Adrián Sánchez-Carmona. A mobile code
bundle extension for application-defined routing in delay and disruption tolerant networking.
Computer Networks, 87:59–77, 2015.
[4] Carlos Borrego, Marica Amadeo, Antonella Molinaro, and Rutvij H Jhaveri. Privacy-preserving
forwarding using homomorphic encryption for information-centric wireless ad hoc networks.
IEEE Communications Letters, 23(10):1708–1711, 2019.
[5] Russell J Clark, Evan Zasoski, Jon Olson, Mostafa Ammar, and Ellen Zegura. D-book: A mobile
social networking application for delay tolerant networks. In Proceedings of the Third ACM
Workshop on Challenged Networks, pp. 113–116, 2008.
[6] Carlos Borrego, Joan Borrell, and Sergi Robles. Hey, influencer! Message delivery to social cen-
tral nodes in social opportunistic networks. Computer Communications, 137:81–91, 2019.
[7] Carlos Borrego, Sergio Castillo, and Sergi Robles. Striving for sensing: Taming your mobile code
to share a robot sensor network. Information Sciences, 277:338–357, 2014.
[8] Carlos Borrego, Gerard Garcia, and Sergi Robles. Softwarecast: A code-based delivery manycast
scheme in heterogeneous and opportunistic ad hoc networks. Ad Hoc Networks, 55:72–86,
2017.
[9] Naercio Magaia, Carlos Borrego, Paulo Pereira, and Miguel Correia. Privo: A privacy-preserving
opportunistic routing protocol for delay tolerant networks. In 2017 IFIP Networking Conference
(IFIP Networking) and Workshops, pp. 1–9. IEEE, 2017.
[10] Adrián Sánchez-Carmona, Sergi Robles, and Carlos Borrego. Privhab+: A secure geographic
routing protocol for DTN. Computer Communications, 78:56–73, 2016.
[11] J Schovancova, S Campana, DC Van Der Ster, J Elmsheuser, T Kouba, R Medrano Llamas, G
Sciacca, F Legger, E Magradze, G Negri, et al. Atlas distributed computing automation. Technical
report, ATL-COM-SOFT-2012-157, 2012.
[12] Erwin Laure, A Edlund, F Pacini, P Buncic, M Barroso, A Di Meglio, F Prelz, A Frohner, O
Mulmo, A Krenek, et al. Programming the grid with glite. Technical report, 2006.
 Opportunistic Emergency Scenarios  35

[13] Julia Andreeva, C Borrego Iglesias, S Campana, A Di Girolamo, I Dzhunov, X Espinal Curull, S
Gayazov, E Magradze, MM Nowotka, L Rinaldi, et al. Automating atlas computing operations
using the site status board. Journal of Physics: Conference Series, vol. 396:1–6, 2012.
[14] Paulo F Jarschel, Jin H Kim, Louis Biadala, Maxime Berthe, Yannick Lambert, Richard M
Osgood, Gilles Patriarche, Bruno Grandidier, and Jimmy Xu. Single-electron tunneling pbs/inp
neuromorphic computing building blocks. arXiv preprint arXiv:1908.08602, 2019.
[15] Carlos Borrego, Adrián Sánchez-Carmona, Zhiyuan Li, and Sergi Robles. Explore and wait:
A composite routing-delivery scheme for relative profile-casting in opportunistic networks.
Computer Networks, 123:51–63, 2017.
[16] Aruna Balasubramanian, Brian Levine, and Arun Venkataramani. DTN routing as a resource
allocation problem. In Proceedings of the 2007 Conference on Applications, Technologies,
Architectures, and Protocols for Computer Communications, pp. 373–384, 2007.
[17] Moazam Ali and Susmit Bagchi. Probabilistic normed load monitoring in large scale distributed
systems using mobile agents. Future Generation Computer Systems, 96:148–167, 2019.
[18] Tatiana Pereira Filgueiras, Leonardo M Rodrigues, Luciana de Oliveira Rech, Luciana Moreira
Sá de Souza, and Hylson Vescovi Netto. Rt-jade: A preemptive real-time scheduling middle-
ware for mobile agents. Concurrency and Computation: Practice and Experience, 31(13):e5061,
2019.
[19] Akanksha Ahuja, Sharad Agarwal, and David Lange. Growth and evolution CMS offline com-
puting from run 1 to HL-LHC. 2020.
[20] K Scott, S Burleigh, et al. RFC 5050: bundle protocol specification. IRTF DTN Research Group,
2007.
[21] Dario Bruneo, Salvatore Distefano, Maurizio Giacobbe, Antonino Longo Minnolo, Francesco
Longo, Giovanni Merlino, Davide Mulfari, Alfonso Panarello, Giuseppe Patanè, Antonio
Puliafito, et al. An IoT service ecosystem for smart cities: The# smartme project. Internet of
Things, 5:12–33, 2019.
[22] Xinxing Zhao, Chandra Sekar Veerappan, Peter Loh, and Caleb Wee. Towards cross-platform
detection of cyberattacks using micro-agents. International Journal of Cyber-Security and Digital
Forensics, 8(2):108–120, 2019.
[23] Vinton Cerf, Scott Burleigh, Adrian Hooke, Leigh Torgerson, Robert Durst, Keith Scott, Kevin
Fall, and Howard Weiss. Rfc 4838, delay-tolerant networking architecture. IRTF DTN Research
Group, 2(4):6, 2007.
[24] Abraham Martín-Campillo, Jon Crowcroft, Eiko Yoneki, and Ramon Martí. Evaluating
opportunistic networks in disaster scenarios. Journal of Network and Computer Applications,
36(2):870–880, 2013.
[25] Carlos Borrego, Joan Borrell, and Sergi Robles. Efficient broadcast in opportunistic networks
using optimal stopping theory. Ad Hoc Networks, 88:5–17, 2019.
Chapter 3

Reactive and Proactive Routing Strategies in

Mobile Ad Hoc Network
Md. Ibrahim Talukdar and Md. Sharif Hossen
Comilla University, Cumilla, Bangladesh

CONTENTS

3.1 Introduction and Related Works................................................................................ 37

3.2 State of the Routing Strategies with Mobility Models (MM)..................................... 38
3.2.1 Protocols under Investigation .........................................................................39
3.2.1.1 Ad Hoc On-Demand Distance Vector (AODV) ................................. 39
3.2.1.2 Dynamic Source Routing (DSR)........................................................ 41
3.2.1.3 Destination-Sequenced Distance-Vector (DSDV)............................... 41
3.2.2 Mobility Models (MM)................................................................................... 42
3.2.2.1 Random Walk (RW)......................................................................... 42
3.2.2.2 Random Waypoint (RWP) ................................................................ 42
3.2.2.3 Random Direction (RD)................................................................... 42
3.3 Methodology and Environment Setting...................................................................... 43
3.3.1 Simulation Tool............................................................................................... 43
3.3.2 Random Traffic and Mobility Generation....................................................... 43
3.3.3 Running the NAM File and Data Sending....................................................... 43
3.3.4 Simulation Environment Setup........................................................................ 43
3.4 Result and Discussion................................................................................................ 43
3.4.1 Evaluation Metrics.......................................................................................... 43
3.4.2 Performance Analysis through Varying the Number of Nodes ........................45
3.4.3 Performance Analysis through Varying the Packet Sizes.................................. 46
3.4.4 Performance Analysis through Varying the Data Rates.................................... 47
3.4.5 Performance Analysis through Varying the Speed of Nodes............................. 50
3.5 Conclusion ................................................................................................................52
Acknowledgements............................................................................................................52
References.......................................................................................................................... 52

3.1 INTRODUCTION AND RELATED WORKS

Wireless networks are classified as infrastructure-based networks with a central access point,
and ad hoc networks with no access point. Mobile ad hoc networks (MANETs) [1, 2] are
wireless in nature with no infrastructure, where each node/device may act as a source or a
router or a destination, etc., and the bridge of information forwards packets for nodes that
are not in the transmission range. These nodes or devices can have different speeds, transmis-
sion ranges, data rates, and packet sizes. Some of the characteristics [3] of the ad-hoc network
are autonomous, dynamic topology, multi-hop, etc. These networks are also constrained to
the transmission ranges, packet losses, security, Quality of Service (QoS), hidden terminal

37
38  Opportunistic Networks

problems, etc. MANET protocols can be categorized into three schemes, such as proactive,
reactive, and hybrid (a combination of both). Regardless of various protocols, only Ad Hoc
On-Demand Distance Vector (AODV), Destination-Sequenced Distance-Vector Routing
(DSDV), and Dynamic Source Routing (DSR) are examined here.
A myriad of works is performed on an ad hoc network. Authors [4] examine the reactive
protocols in [5]; the proactive and reactive protocols are analyzed using NS-3. The authors
[6] evaluate MANET protocols using OPNET. Moreover, the energy efficiency of the AODV
protocol is analyzed in [7]. Special kinds of ad hoc networks, e.g., Vehicular Ad hoc Networks
(VANETs), are more realistic in intelligent transport systems, which are more popular today.
In [8, 9], the routing protocols are evaluated for the VANET environments.
Ad hoc networks have a variety of application areas [10] but are not limited to vehicle
communication, distributed computing, earthquake/tsunami-hit areas, sensor networks, mili-
tary applications, multimedia applications, space exploration, tracking of rare animals, bio-
logical detection, undersea operations, Personal Area Network (PAN), Bluetooth, etc. Hence,
MANETs have emerged as a prime research area. Diverse applications of these networks in
many real-time scenarios stimulate us to judge the protocols of MANETs. There are many
papers on the performance measurement of ad hoc with the basis of different metrics and
variations. The key motivation behind this chapter is the better understanding of protocols
where they are applicable or not with challenged and crucial variants like mobile nodes,
velocity, data rates, and packet sizes. In this chapter, MANET protocols like AODV [11, 12],
DSDV [13], and DSR [14] are examined to select the best method for establishing an emer-
gency communication when needed. Here, the significance of ad hoc routing protocols are
examined in terms of the QoS parameters like packet delivery, delay, routing overhead, and
throughput with varying some parameters such as the number of nodes, packet sizes, data
rates, and the velocity of nodes for random simulation area with random waypoint mobility
[15]. In the remaining sections, Section 3.2 describes the ad hoc routing protocols with the
mobility models; Section 3.3 briefly discusses the methodology and the simulation environ-
ment; Section 3.4 depicts the QoS parameters and the graphical results with decent analysis;
Finally, Section 3.5 recapitulates the chapter.

3.2 STATE OF THE ROUTING STRATEGIES WITH MOBILITY MODELS (MM)

Routing is a very crucial part for MANETs. Basically, the mobile ad hoc networks are com-
posed of two routing protocols, namely, proactive and reactive. Moreover, the amalgamation
of both introduces the hybrid routing protocol (Figure 3.1).

Figure 3.1  Routing protocols in MANET.

 Reactive and Proactive Routing Strategies  39

(a) Proactive: The proactive [16] protocols follow a table-driven approach wherein all
the nodes endlessly search for routing information within a network. In this protocol,
each node retains one or more routing table representing the entire network topology.
Hence, the routing table maintenance is difficult for a large topology. These protocols
require more overhead, which consumes more bandwidths. Some of these protocols
are derived from the link-state [17] routings. Optimized Link State Routing (OLSR),
DSDV, etc., are proactive routing protocols in ad hoc networks.
(b) Reactive: The reactive protocols, also termed as on-demand routing protocols, do not
keep track of the routing information until any communication takes place. In the reac-
tive [18] approach, the routing information is accumulated as per need, and the route
determination relays on forwarding the route queries all over the network. The routes
are uncovered by flooding the route request. AODV, Area Border Routing (ABR) [19],
DSR, Temporally Ordered Routing (TORA) [20], etc., are reactive protocols.

3.2.1 Protocols under Investigation

This subsection precisely describes the considered protocols under investigation in this
chapter.

3.2.1.1 Ad Hoc On-Demand Distance Vector (AODV)

AODV discovers the route on user demand fashion, that is, it starts searching for a route
during the transmission of any packet to the destination. It discovers a route, generates the
messages, and maintains the established route. In AODV, no overhead packet is required if
the path exists from one node to another. It disallows the routing loops by using the sequence
numbers. Besides, it uses a broadcast technique for searching a route and the unicast mecha-
nism for responding from the route. It relays on the DSDV and the DSR wherein the infor-
mation on the sequence number and the periodic beaconing of the DSDV and the route
finding technique of the DSR are used. However, the DSR and the AODV have the following
dissimilarities:

(i) The DSR conveys the complete routing information of packets, whereas the AODV
keeps only the destination address which implies that the AODV requires less routing
overheads than the DSR.
(ii) The route response methods of the DSR convey the address of every node along the
path, whereas only the sequence number and the destination IP addresses are carried
by the AODV route reply policy.

AODV protocol follows two phases as depicted in the following.

(a) Route Discovery of AODV: To understand this process, suppose the source node 1
tries to communicate with the destination node 7 where the origin launches a route
discovery through broadcasting the route request with the inclusion of source ID (Sid),
the destination ID (Did), the source sequence (SSeq), the destination sequence (DSeq),
time to live (TTL), etc. to all its neighbors (Figure 3.2).
The neighboring nodes compare the DSeq of Route Request (RREQ) with their
corresponding arrival in route cache. For greater DSeq of RREQ, the neighbor node
responds to the source with an RREP carrying the path to the destination. Generally,
a node (say, node 3) having a route to the destination (node 7) in its cache with higher
40  Opportunistic Networks

DSeq as compared to RREQ sends a Route Reply (RREP) back to the source. Thus, the
route 1-3-6-7 is kept in node 1. Moreover, the destination node also onwards an RREP
back to the origin.
(b) Route Maintenance of AODV: In the route maintenance mechanism of AODV, when
a broken link is detected, the origin and the endpoint are notified by disseminating an
RERR like as the DSR scenario. During the broken link between two nodes (say, node
3 and 5) as depicted in Figure 3.3 on the route 1-3-5-7, both nodes will send an RERR
to the source and the endpoint. The major point of AODV over DSR and AODV is that
it reduces routing overhead through avoiding source routing.

Figure 3.2  Route discovery of AODV.

Figure 3.3  Route Maintenance of AODV.

 Reactive and Proactive Routing Strategies  41

3.2.1.2 Dynamic Source Routing (DSR)

DSR routing uses the idea of source routing wherein the sender has the entire node-to-node
path to the destination. In DSR, the routing information is not needed to be maintained at
every host. It has two phases, path finding and route maintenance, which work jointly to
discover and support source routes to random destinations in the MANET. When the des-
tination route is unknown to a node, it uses a flooding-based route discovery strategy with
RREQ where each one accepts RREQ and resends it to the neighboring nodes until the des-
tination, D, is found. Whenever RREQ reaches to D, D responds to the RREQ with an RREP
packet to the origin by traversing this path backwardly. However, it is not suited for larger
networks, since it consumes most of the bandwidth of the networks [21]. The DSR also fol-
lows two phases like the AODV.

(a) Route Discovery of DSR: In the DSR route discovery phase, the source node, say, node
1, tries to communicate with the destination node (say node 7), wherein the source
starts a route discovery through broadcasting an RREQ to its adjacent nodes, i.e., 2,
3, and 4. Similarly, the intermediate nodes follow the same process. After reaching at
node 7, it adds its own address and reverses the route to the destination using unicast
fashion. The destination node accepts the first received route from a node as the best
path and then caches the other nodes (Figure 3.4).
(b) Route Maintenance of DSR: The route maintenance is required during the broken link
between nodes 6 and 7. The damaged link can be detected by either passive or active
monitoring of the link to a node. In this case, an intermediate node sends a route error
packet to the origin, and then the source reinitiates the path discovery (Figure 3.5).

3.2.1.3 Destination-Sequenced Distance-Vector (DSDV)

The Bellman-Ford approach is utilized in the DSDV with some amelioration. Also, the DSDV,
a proactive protocol, is derived from the routing information protocol (RIP) and placed into
an ad hoc routing. It selects the best path to the destination using the distance vector short-
est path. It adds new attributes and sequence numbers to the routing table, and using this

Figure 3.4  Route Discovery of DSR.

42  Opportunistic Networks

Figure 3.5  Route Maintenance of DSR.

sequence, it handles the infinite loop problems. Due to the large overhead, it consumes more
network resources and battery power. This protocol is suitable for a small network and does
not support multicasting [22].

3.2.2 Mobility Models (MM)

There are lots of MMs, e.g., Random Walk, Random Direction, Random Waypoint (RWP),
etc., to move nodes independently within the simulation area. Among these models, the
RWP is chosen to capture the performance of the networks. It adds the concept of pause
times in a node before changing its speed and direction. Without pause time, it acts like
Random Walk [23].

3.2.2.1 Random Walk (RW)

In RW, the nodes move in an unpredictable fashion. Here, each node conveys random speed
and direction from the predefined ranges until it reaches the destination. Whenever any node
reaches the destination, it receives new speed and direction from that range. In RW, the nodes
move freely without any restriction, and no tracking records of the nodes are kept.

3.2.2.2 Random Waypoint (RWP)

The working principle of the RWP is very similar to the RW mobility, except it adds the
notion of pause time while moving nodes. With zero pause time, the RWP and the RW are the
same. Very often the RWP faces non-uniform node distribution problems. However, by using
a smooth RWP (SRWP) technique with the concept of temporal dependency, it improves the
performance of the RWP.

3.2.2.3 Random Direction (RD)

RD overcomes the problem that arises in the RWP. In the RD mobility, a mobile node pro-
ceeds in a specific direction until it touches the destination border of the simulation area with
a fixed speed. It suggests an even distribution of the nodes in the specified simulation region.
 Reactive and Proactive Routing Strategies  43

3.3 METHODOLOGY AND ENVIRONMENT SETTING

The simulation methodology and environment settings are described briefly in this section.

3.3.1 Simulation Tool
The network simulator 2, called NS-2, along with Linux, is used as a simulation tool. It
uses an object tool command language shell (OTCL) as the user interface which allows the
scripts written in TCL format to be executed. It is a discrete event simulator along with the
object-oriented concepts and used by the ad hoc networking community and researchers to
determine the performance of routings. It uses two languages: (1) C++, which runs in the
back end, and (2) tool command language (TCL), which runs in the front end [24]. Moreover,
the components of NS-2 include NS, NAM (network animator), pre-processing, and post-
analysis for the simulator, visualization of NS output, handwritten TCL, and the trace analy-
sis using AWK scripts.

3.3.2 Random Traffic and Mobility Generation

The CBRGEN tool under “~ns/indep-utils/cmu-scen-gen” is found to generate the random
traffic among the nodes using TCP or CBR connection. Moreover, the SETDEST tool under
“~ns/indep-utils/cmu-scen-gen/setdest/” is used to produce the traces of the nodes by the
random movement with the velocity of the nodes to any location (not fixed) within the
considered wireless region. The node’s mobility is distributed by a random waypoint fash-
ion [24]. However, the traffic connections and nodes’ mobility can be made manually for a
small network.

3.3.3 Running the NAM File and Data Sending

After running the TCL script for a specified network environment, the NAM and the trace
files are generated. Then, using AWK scripts, PDR, delay, t-put (throughput), etc., are exam-
ined. The NAM is used for tracking real-world traces. Here is the scenario of running the
NAM file (Figure 3.6).

3.3.4 Simulation Environment Setup

In this research, the wireless network environment is constructed for moving the nodes’ CBR
traffic patterns, and the RWP mobility models with specified simulation area, time, etc. The
wireless channel with the random waypoint is considered as the mobility model. Table 3.1
shows the general simulation parameters in detail.
Table 3.2 shows the variations of the mobile nodes, packets, data rates, and velocity of
nodes that are used in environment simulation.

3.4 RESULT AND DISCUSSION

3.4.1 Evaluation Metrics
The following metrics are examined to analyze the performance of protocols.
44  Opportunistic Networks

Figure 3.6  The NAM file and data sending.

Table 3.1  General parameters for simulation.

Parameters Values

Routing Protocols AODV, DSDV, and DSR

Mobility Mode Two Ray Ground
Antenna Omni Antenna
Simulation Time 160 sec
Channel Wireless Channel
Mobility Model Random Way Point
Simulation Area 1100 × 800
Traffic CBR/UDP
Packet Size 1500 Bytes
MAC MAC/802-11
Mobile Nodes 30
Mobility Speed 6 m/s
Data Rates 0.1 Mbps
Performance Metrics Packet Delivery Ratio, End-to-end Delay, Normalized Routing
Overhead, Throughput
Simulator NS 2.35

(i) Packet delivery ratio (PDR): PDR, a vital issue of a network, can be expressed as the
number of packets received at the destination over the packets sent from the source.
Higher packet delivery ensures better performance [25]. It is determined as
Total packet received
PDR   100%
Total packet sent
 Reactive and Proactive Routing Strategies  45

Table 3.2  Variation of parameters.

Parameters Values

Mobile Nodes 20, 30, 40, 50, and 60

Packet Sizes 512, 1000, 1500, 1700, and 200 Bytes
Data Rates 0.1, 2, 5, 10, and 15 Mbps
Speeds 6, 10, 20, 25, and 35 m/s

(ii) End-to-end delay: It is the time between the packet generated by the source node and
the packet received by the destination node. It can be expressed as follows at millisec-
ond (ms) [26].
n
Received time  Sent time

End - to- end Delay   Numbers of packets received
i 1

(iii) Overhead: The scheme defines how many redundant packets are relayed to convey
one packet. In other words, it can be expressed as follows.

Total data packets sent

Normalized Routing Overhead =
Total data packets received

(iv) Throughput (T-put): The average throughput, in Mbps or Kbps, refers to how fast a
packet can be sent from one to another. High t-put is desirable for every network. It is
directly proportional to PDR.

3.4.2 Performance Analysis through Varying the Number of Nodes

For varying the number of nodes, it is clear from Figure 3.7 that the AODV protocol achieves
slightly higher delivery as compared to the DSR. It is also noticeable that the delivery
probability is likely in increasing order with the increase of the mobile nodes. Among the
protocols, the DSDV provides the least performance in case of the delivery.

Figure 3.7  PDR with varying the number of nodes.

46  Opportunistic Networks

From Figure 3.8, it is obvious that the DSDV outperforms the AODV and the DSR with
varying the number of nodes as it services few messages to the destination. Moreover, the
AODV approximately exhibits lower delay than the DSR.
Figure 3.9 depicts that with the increase of the number of mobile nodes, the normalized
routing overhead is increasing for the AODV, and the DSDV where the DSDV shows the
highest overhead value and the DSR exhibits the least overhead (approximately zero).
Again, for varying the number of nodes, it is seen that the AODV ensures the best through-
put and DSDV performs the least among the protocols shown in Figure 3.10. Moreover, it is
blatant that the throughput is proportional to the number of mobile nodes.

3.4.3 Performance Analysis through Varying the Packet Sizes

Figure 3.11 depicts that the AODV presents a higher delivery ratio with respect to the packet
size as it does for the nodes’ variation. Notice also that the DSDV delivers lower packets than
the AODV and the DSR.

Figure 3.8  Delay with varying the number of nodes.

Figure 3.9  Overhead with varying the number of nodes.

 Reactive and Proactive Routing Strategies  47

Figure 3.10  T-put with varying the number of nodes.

Figure 3.11  PDR with varying the packet sizes.

Figure 3.12 depicts that the average delay is fluctuating for varying the size of packets. The
average delay is extremely high for the DSR with respect to the packets’ variation because the
DSR collects and maintains all the possible routes to a destination node.
The DSDV consumes a very large number of extra packets for data delivery as illustrated in
Figure 3.13. It shows the highest overhead value among all the protocols. As with the node’s
variation, the variation of packet size also shows that the DSR has the least overhead value
(approximately zero).
The AODV routing achieves a slightly higher throughput value than the DSR. Here, the
DSDV shows the lowest throughput compared to the AODV and the DSR. With a packet
size of 512 bytes, all the protocols reach their maximum gain. However, the throughput is
decreasing with the increase of packet sizes for all the protocols, as shown in Figure 3.14.

3.4.4 Performance Analysis through Varying the Data Rates

From Figure 3.15, it seems that the values overlap with each other. Actually, the PDR is at
maximum for the AODV, lowest for the DSDV, and medium for the DSR. Additionally, it
is blatant that the packet delivery is decreasing with the increase of data rates for all the
48  Opportunistic Networks

Figure 3.12  Delay with varying the packet sizes.

Figure 3.13  Overhead with varying the packet sizes.

Figure 3.14  T-put with varying the packet sizes.

 Reactive and Proactive Routing Strategies  49

protocols. However, the AODV outperforms the DSDV and the DSR in terms of the packet
delivery with the alteration of the data rates, as shown in Figure 3.15.
Figure 3.16 indicates that the average delay is increasing overall with the increase of the
data rates. DSR badly exhibits high latency due to complete the route maintenance from the
source to the destination. Proactive protocol (i.e. DSDV) shows the lowest delay.
The AODV routing exhibits lower overhead than the DSDV when the data rates are within
0.1 and 2 MB. With the increase of data rates, the overhead is improving for the DSDV than
the AODV. The DSR is not affected by the data rates and exhibits constant, and approxi-
mately zero overhead (Figure 3.17).
Figure 3.18 depicts that the AODV achieves maximum throughput with the variety of data
rates while the DSDV shows lower t-put. In contrast to the message delivery, the throughput
is increasing for varying the data rates.

Figure 3.15  PDR with varying the data rates.

Figure 3.16  Delay with varying the data rates.

50  Opportunistic Networks

Figure 3.17  Overhead with varying the data rates.

Figure 3.18  T-put with varying the data rates.

3.4.5 Performance Analysis through Varying the Speed of Nodes

Figure 3.19 illustrates that the changes of the nodes’ speed significantly affect the packet
delivery. The AODV shows the maximum packet delivery than others wherein the DSDV
exhibits the least delivery. But PDR is decreasing with the increase in the number of nodes
for reactive protocols.
Figure 3.20 depicts that the delay is fluctuating for the reactive protocols. The average
delay is minimum for DSDV protocol, as are all the variants under consideration.
In all cases (Figures 3.9, 3.13, 3.17, and 3.21), it is obvious that the DSR protocol exhibits
the least (approximately zero) overhead regardless of all the variants. The AODV routing
shows lower overhead compared to the DSDV, as shown in Figure 3.21.
Figure 3.22 depicts that the throughput of the AODV routing is significantly higher than
the DSR and the DSDV, as discussed previously, for varying the number of nodes, the packet
sizes, and the data rates. The DSDV shows the least throughput, like other cases.
 Reactive and Proactive Routing Strategies  51

Figure 3.19  PDR with varying the speed of nodes.

Figure 3.20  Delay with varying the speed of nodes.

Figure 3.21  Overhead with varying the speed of nodes.

52  Opportunistic Networks

Figure 3.22  T-put with varying the speed of nodes.

3.5 CONCLUSION

To ensure which protocol is best for constructing mobile networks in an emergency situation,
this research is performed by comparing the performance of proactive and reactive tech-
niques using NS-2. This chapter reveals the performance of AODV, DSDV, and DSR for vary-
ing the number of nodes, packet sizes, data rates, and node’s speeds within a specified area
on the basis of packet delivery, delay, overhead, and throughput. The pictorial analysis shows
that the AODV protocol outperforms the DSDV and the DSR in terms of delivery with all the
variants. Also, it ensures better throughput among all the protocols considered here which
blatantly indicates that the delivery is often changing proportionally with the throughput.
Again, the graphical results depict that the DSDV protocol achieves desirable delay among
the protocols since its delivery is very low compared to the reactive protocols, whereas the
DSR shows higher delay compared to the AODV since its header is larger than the AODV.
Here, the simulation results also show that the DSR experiences no overhead, and the AODV
shows lower overhead compared to the DSDV. In addition, each routing protocol has its pros
and cons and should be used as the application requires. It is rarely possible to name one
protocol as being better than the others in any network under all criteria. Considering the
overall performances, the AODV performs well as compared to other routing approaches for
considered areas. However, the security issues of the routing techniques are not considered
here. The investigation in this chapter has been done under no attack, so there is a possibility
of changing the performance of the routing approaches during the security issues.

ACKNOWLEDGEMENTS

The authors would like to thank the ICT Division of the Bangladesh Government for award-
ing a research fellowship to Md. Ibrahim Talukdar for this research.

REFERENCES

[1] Talukdar, M. I. (2021). Performance Improvements of AODV by Black Hole Attack Detection
Using IDS and Digital Signature. Wireless Communications and Mobile Computing, 2021(1),
pp. 1–13.
 Reactive and Proactive Routing Strategies  53

[2] Salam, T. (2020, January 21). Performance Analysis on Homogeneous LEACH and EAMMH
Protocols in Wireless Sensor Network. Wireless Personal Communications. doi:10.1007/
s11277-020-07185-6.
[3] Venkatesan, T. P. (2014, April). Overview of Proactive Routing Protocols in MANET. Fourth
International Conference on Communication Systems and Network Technologies, pp. 173–177.
doi:10.1109/CSNT.2014.42.
[4] Mohapatra, R. K. (2018, October). Performance Analysis of Reactive Routing Protocols in
MANET under CBR Traffic using NS2. International Conference on Advances in Computing,
Communication Control and Networking (ICACCCN), pp. 352–356. doi:10.1109/
ICACCCN.2018.8748851.
[5] Raman Chandan, R. (2018). Performance Evaluation of AODV, DSDV, OLSR Routing Protocols
using NS-3 Simulator. International Journal of Computer Network and Information Security,
10(7), pp. 59–65. doi:10.5815/ijcnis.2018.07.07.
[6] Aneiba, A., Melad, M. (2016). Performance Evaluation of AODV, DSR, OLSR, and GRP
MANET Routing Protocols Using OPNET. International Journal of Future Computer and
Communication, 5(1), pp. 57–60.
[7] Anand, M. (2019). Efficient Energy Optimization in Mobile Ad Hoc Network (MANET) Using
Better Quality AODV Protocol. Cluster Computing, 22(S5), pp. 12681–12687. doi:10.1007/
s10586-018-1721-2.
[8] Ur Rehman, M. (2018). Performance and Execution Evaluation of VANETs Routing Protocols
in Different Scenarios. EAI Endorsed Transactions on Energy Web, 5(17), pp. 154458–154455.
doi:10.4108/eai.10-4-2018.154458.
[9] Poonia, R. C. (2018). A Performance Evaluation of Routing Protocols for Vehicular Ad Hoc
Networks with Swarm Intelligence. International Journal of System Assurance Engineering and
Management, 9(4), pp. 830–835. doi:10.1007/s13198-017-0661-1.
[10] Hossen, M. S., Rahim, M. S. (2019). Analysis of Delay-Tolerant Routing Protocols Using the
Impact of Mobility Models. Journal of Scalable Computing: Practice and Experience, 20(1), pp.
39–48.
[11] Hossen, M. S., Rahim, M. S. (2015). On the Performance of Delay-Tolerant Routing Protocols
in Intermittently Connected Mobile Networks. Rajshahi University Journal of Science &
Engineering, 43, pp. 29–38.
[12] Swidan, A. (2016). Mobility and Direction Aware Ad-Hoc on Demand Distance Vector Routing
Protocol. Procedia Computer Science, 94, pp. 49–56. doi:10.1016/j.procs.2016.08.011.
[13] Wang, J. (2010). A Secure Destination-Sequenced Distance-Vector Routing Protocol for Ad Hoc
Networks. Journal of Networks, 5(8), p. 942.
[14] Suriya, C. (2014). A Study on Dynamic Source Routing In Ad Hoc Wireless Networks.
International Journal of Advanced Networking and Applications, 5(5), p. 2046.
[15] Zhong, Y. (2003). Dynamic Source Routing Protocol for Wireless Ad Hoc Networks in Special
Scenario Using Location Information. International Conference on Communication Technology
Proceedings, Vol. 2, pp. 1287–1290. doi:10.1109/ICCT.2003.1209766.
[16] Mohseni, S. (2010, April). Comparative Review Study of Reactive and Proactive Routing Protocols
in MANETs. 4th IEEE International Conference on Digital Ecosystems and Technologies, pp.
304–309. doi:10.1109/DEST.2010.5610631.
[17] Hong, X. (2002). Scalable Routing Protocols for Mobile Ad Hoc Networks. IEEE Network,
16(4), pp. 11–21. doi:10.1109/MNET.2002.1020231.
[18] Araghi, T. K. (2013, September). Performance Analysis in Reactive Routing Protocols in Wireless
Mobile Ad Hoc Networks Using DSR, AODV and AOMDV. International Conference on
Informatics and Creative Multimedia, pp. 81–84. doi:10.1109/ICICM.2013.62.
[19] Gupta, S. (2015). Performance Comparison of Proactive Routing Protocols: OLSR, DSDV, WRP.
International Journal of Advanced Research in Computer Science, 6(8) pp. 73–77.
[20] Mafirabadza, C. (2017). Efficient Power Aware AODV Routing Protocol for MANET. Wireless
Personal Communications, 97(4), pp. 5707–5717. doi:10.1007/s11277-017-4804-0.
[21] Das, S. (2000). Performance Comparison of Two On-Demand Routing Protocols for Ad Hoc
Networks. Proceedings IEEE INFOCOM 2000. Conference on Computer Communications.
54  Opportunistic Networks

Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat.
No.00CH37064), Vol. 1, pp. 3–12. doi:10.1109/INFCOM.2000.832168.
[22] Madhusrhee, B. (2016). Performance Analysis of AODV, DSDV and AOMDV Using WiMAX in
NS-2. Computational Methods in Social Sciences, 4(1), pp. 22–28.
[23] Hossen, S. (2016, January). Impact of Mobile Nodes for Few Mobility Models on Delay-Tolerant
Network Routing Protocols. International Conference on Networking Systems and Security
(NSysS), pp. 1–6. doi:10.1109/NSysS.2016.7400704.
[24] Issariyakul, T. (2011). An Introduction to Network Simulator NS2. New York, NY: Springer.
[25] Hossen, M. S. (2019). DTN Routing Protocols on Two Distinct Geographical Regions in an
Opportunistic Network: An Analysis. Wireless Personal Communications, 108, pp. 839–851.
doi:10.1007/s11277-019-06431-w.8.
[26] Talukdar, M. I., Hossen, M. S. (2019). Selecting Mobility Model and Routing Protocol for
Establishing Emergency Communication in a Congested City for Delay-Tolerant Network.
International Journal of Sensor Networks and Data Communications, 8(1), pp. 1–9.
doi:10.4172/2090-4886.1000163.
Chapter 4

Secure Hierarchical Infrastructure-Based

Privacy Preservation Authentication
Scheme in Vehicular Ad Hoc Networks
Mariya Ouaissa, Mariyam Ouaissa and Meriem Houmer
Moulay Ismail University, Meknes, Morocco

CONTENTS

4.1 Introduction............................................................................................................... 55
4.2 Related Works........................................................................................................... 56
4.3 Background............................................................................................................... 58
4.3.1 System Architecture of VANET.................................................................... 58
4.3.2 Security Requirements in VANET................................................................ 59
4.3.2.1 Authentication................................................................................ 60
4.3.2.2 Confidentiality................................................................................ 60
4.3.2.3 Integrity.......................................................................................... 61
4.3.2.4 Availability...................................................................................... 61
4.3.2.5 Non-repudiation............................................................................. 62
4.3.2.6 Privacy Preserving........................................................................... 62
4.4 Preliminaries.............................................................................................................. 62
4.4.1 Elliptic Curve Diffie-Hellman (ECDH)........................................................ 62
4.4.2 Edwards-curve Digital Signature (EdDSA)................................................... 63
4.5 Proposed Scheme....................................................................................................... 64
4.5.1 Protocol Description.................................................................................... 64
4.5.2 Registration Phase....................................................................................... 65
4.5.3 Authentication and Communication between OBUi and RSUi..................... 67
4.5.4 Communication between Vehicles................................................................ 69
4.5.5 Revocation.................................................................................................. 69
4.6 Security Analysis........................................................................................................ 70
4.6.1 Analysis of Security Requirements............................................................... 70
4.6.2 Formal Verification...................................................................................... 71
4.7 Performance Evaluation............................................................................................. 72
4.7.1 Comparison of Security Performance........................................................... 73
4.7.2 Computation Overhead............................................................................... 73
4.8 Conclusion................................................................................................................. 75
References.......................................................................................................................... 78

4.1 INTRODUCTION

Wireless networks have experienced remarkable progress in recent years; they are undeniable
today. Their appearance and the advancement of communication and information technolo-
gies are giving rise to the so-called Intelligent Transport Systems (ITS). The principal aim of
these systems is to make the road more efficient. Moreover, one of the main strengths of ITS

55
56  Opportunistic Networks

is to enable a level of cooperation between participants in the road network by equipping

vehicles with wireless communication equipment. This type of wireless network refers to
vehicular networks [1].
Vehicular Ad hoc Networks (VANETs) [2] are a new form of Mobile Ad hoc Networks
(MANETs) that aim to provide communications between vehicles or with infrastructure
located at roadside. These networks are described as a dynamic topology according to the
addition or departure of a vehicle from the network. In these networks, vehicles are equipped
with wireless short and medium-range communication. Actually, the vehicles can com-
municate with each other with two methods, either Vehicle to Vehicle (V2V) or Vehicle to
Infrastructure (V2I) where the vehicle communicates with the equipment next to the road
named Road Side Unit (RSU) [3]. VANETs are used to meet the communication needs applied
to transport networks to improve driving and road safety for road users.
In VANET, vehicles exchange messages and communicate with each other in a wireless
environment. This situation can give rise to internal or external security attacks which can
have the objective of rendering the network non-functional, of causing an accident. For
that reason, the preservation of the security of information exchanged between vehicles is a
crucial necessity. Communication must go through the analysis of the potential of security
threats, and the design of a robust architecture capable of dealing with these threats. In this
context, the implementation of vehicular networks requires an effective security mechanism
in order to satisfy security requirements such as authentication, integrity, and the privacy
preservation of a user [4].
This chapter presents a hierarchical authentication protocol with the aim of ensuring basic
security requirements such as integrity, confidentiality, non-repudiation, and availability, as
well as the preservation privacy of users by using an identifier or a real identifier. First, our
solution allows vehicles and RSUs to authenticate with the certification authority so that they
are legal entities in the network. Then a mutual authentication between vehicles and RSUs
makes it possible to minimize access to malicious entities in the network.
The proposed protocol combines the symmetrical and asymmetrical approaches, where
the symmetrical approach used the Elliptic Curve Diffie Hellman (ECDH) algorithm for
reliable key exchange that is implemented in order to create and share the secret key, whose
objective is to ensure the security of the exchange of the parameters for the asymmetric
system (Private Key/Public Key), authentication packets, and vehicle certificate using the
Edwards-curve Digital Signature Algorithm (EdDSA). In addition, the creation of sub-lists of
revoked certificates based on vehicle type makes it possible to minimize the response time by
looking for a certificate whether it is revoked or not.
The remainder of this chapter is organized as follows. The next section details the related
works. Section 4.3 presents the background, including communication architecture and
security requirements in VANETs. Section 4.4 describes several preliminaries used in our
solution. In Section 4.5, we propose a secure hierarchical infrastructure-based privacy
preservation authentication scheme in VANET. The security of the designed protocol was
checked by the verification tool Automated Validation of Internet Security Protocols and
Applications (AVISPA), in Section 4.6. Section 4.7 evaluates and analyzes the perfor-
mances of the existing protocols as well as our proposition. We draw our conclusion in
Section 4.8.

4.2 RELATED WORKS

Authentication represents an essential cryptographic mechanism that provides confidence

between vehicles and infrastructures in vehicle ad hoc networks. However, an improved
 Secure Hierarchical Infrastructure  57

authentication process will effectively detect malicious nodes and then maintain VANET
security. Therefore, various authentication mechanisms that ensure protected communication
in the network were suggested. In this section, we discuss them briefly.
The basic idea of the technique based on anonymous certificates is given by Raya and
Hubaux [5]. The authors use anonymous certificates (e.g. pseudonyms) to hide the real iden-
tity of users. The anonymous certificate does not include any information about the real iden-
tities of the users, but privacy may be violated because the messages contain an exchanged
key which makes it possible to track the vehicle’s true identity.
The group signature is an alternative to achieve security and preserve privacy in VANETs.
In this technique, a group manager is responsible for managing the group. The members may
enter or exit the group dynamically. Upon registering and joining a group, a member can
sign anonymously on the behalf of the group, and the recipient uses the public key to vali-
date the signature but will never know who sent the packet. However, there are exceptional
cases where the group manager may reveal the identity of a sender of any group signature.
The group signing approach has emerged to overcome the disadvantage of the anonymous
certificate technique. The first protocol of this technique to be implemented within vehicle
networks is the Group Signature ID-based Signature protocol [6].
The authors of [7] aim to guarantee the identification, authentication, non-repudiation,
and integrity of the roadside unit when transmitting messages from RSU to vehicles (I2V).
An identification aggregation will be carried out by several RSUs and without the interven-
tion of a trusted third party. Their algorithm first ensures the identification of RSUs by the
Elliptic Curve Diffie-Hellman’s algorithm, which verifies both that the nearest RSU possesses
the same secret key and that the vehicle authenticates the signed message utilizing the Digital
Signature Elliptic Curve. The ECDH-ECDSA provides a greater degree of protection even if
its aggregation takes roughly 40 ms further than the basic ECDSA method.
Vijayabharathi and Malarchelvi [8] present an Expedite Message Authentication Protocol
(EMAP) based Hash Message Authentication Code (HMAC) for vehicular networks. The
authors propose a new fast process of revocation checking to minimize the computation
process and avoid overhead problems.
A new conception of Public Key Infrastructure (PKI) for the authentication process is pro-
posed in [9]. The authors design an infrastructure PKI-based symmetric encryption in order
to reduce the treatment time and eliminate overhead for authentication.
Das et al. [10] offer a hierarchical protocol to reach the objectives of scalability and cer-
tification in VANET. This protocol built a structure of tree with a hierarchy of Certification
Authority (CA) to operate the VANET. Authors suggest two kinds of nodes; the powerful
nodes are the certifying authorities, and the leaf nodes are vehicles.
The Lightweight Identity Authentication Protocol (LIAP) was proposed in [11] in order to
supply a fast mutual authentication between the roadside unit and the vehicle, and also to
guarantee the vehicle’s conditional privacy. This protocol achieves the transfer authentication
process by employing a secret dynamic session mechanism and avoids the utilization of the
encryption/decryption operations in the roadside unit and the vehicle.
The scheme proposed in [12] constitutes a group of vehicles and RSU through the use of
self-authentication without the need of a certification authority. It uses a Group Key (GK) to
improve the efficiency of certification, and the protocol selects deniable group key agreement
method to avoid attacks into legal vehicles.
The authors in [13] propose a new protocol based on the complexity of two popular math-
ematical problems to deal with the problems existing in previous Mobile Wireless Networks’
(MWNs) handover authentication protocols. Security analysis illustrates that the proposed
protocol is protected from various threats and can satisfy a number of security requirements.
A hierarchical revocable authentication protocol based random oracle model within the
58  Opportunistic Networks

Diffie-Hellman (DH) hypothesis is presented in [14]. The evaluation of the protocol shows
that it saves highly computation overheads and meets the security requirements.
Our proposal also aims to ensure privacy and authentication in a vehicular ad hoc
network. The privacy is preserved by ensuring anonymity. In this study, we use an ECDH
algorithm for reliable key exchange and the EdDSA algorithm to speed up the execution
of the authentication process, especially at the key management level, message signing, and
verification of this signature.

4.3 BACKGROUND

This section introduces the VANET communication system architecture and discusses the
VANET’s safety requirements.

4.3.1 System Architecture of VANET

The vehicular ad hoc network is a highly mobile ad hoc network integrated by ITS in order
to improve traffic efficiency, minimize traffic congestion, avoid accidents, and make easy
access to news, information, and entertainment while driving. However, recent research in
vehicular networks is exploring all aspects of communication. We distinguish three modes
of communication in a VANET, which are vehicle-to-vehicle communication (V2V), vehicle-
to-infrastructure communication (V2I) and infrastructure-to-infrastructure communication
(I2I) [15].

• Vehicle-to-vehicle communication (V2V) is based on a simple inter-vehicle communica-

tion that utilizes the OBUs (On-Board Units) installed on each vehicle to communicate
with each other. This communication can be established without fixed infrastructure
relays.
• Vehicle-to-infrastructure communication (V2I) allows better use of shared resources
and increases the services provided (for example, data exchange, Internet access, remote
diagnostics to repair a vehicle, etc.) thanks to access points deployed at the roadside.
These access points are named RSUs (road side units) and are located in certain critical
sections of the road, such as traffic lights or stop signs, in order to enhance road safety
and traffic efficiency, and also to enjoy driving.
• Infrastructure to infrastructure communication (I2I) provides communication between
RSUs or between RSU and base station. It increases the communication range and con-
nects all vehicles in the network.

As the range of infrastructure is limited, vehicles can be used as relays to extend this dis-
tance and avoid the multiplication of base stations at each corner of the road. Therefore, the
combination of the communication modes (V2V, V2I, and I2I) may achieve a very interest-
ing and economical hybrid communication. In order to establish all these communications,
a VANET consists of three main components (OBU, RSU, and CA) [16] required to provide
communications in the network (Figure 4.1).

• The On-Board Unit (OBU) is a sensor mounted in vehicles. This device provides impor-
tant information to vehicle control units for automatic driving assistance. This on-
board unit is used to exchange information with other OBUs or with road side units
(RSUs). It includes a set of high-tech hardware and software components such as GPS,
 Secure Hierarchical Infrastructure  59

Figure 4.1  Network model.

radar, cameras, various sensors, and others. Its role is to ensure the location, reception,
calculation, storage, and sending data over the network.
• The Road Side Unit (RSU) is an infrastructure placed along roads. Its main role is to
inform nearby vehicles by broadcasting traffic conditions, weather, or specific road con-
ditions (maximum speed, overtaking, etc.). They can also play the role of a base station
by relaying information sent by vehicles.
• The Certification Authority (CA) represents the trusted authority in the VANET. It plays
the role of a server which ensures the security of the various services such as the issuing
of certificates, keys distribution, and the storage of certain data. Moreover, the CA
maintains secure management of the VANET by verifying the vehicle authentication,
the user ID, and the OBU ID in order to prevent any damage to any vehicle.

4.3.2 Security Requirements in VANET

All the constraints weighing on a system need to be understood to satisfy its security
requirements. This step then makes it possible to quantify the security criteria. Regarding
vehicular networks, the implementation of a security policy requires a good understanding
and apprehension of the characteristics and constraints of these networks.
Since vehicular networks are regarded as a subclass of mobile ad hoc networks, they
have the same characteristics, with a few exceptions. In contrast to traditional ad-hoc
networks, VANETs have positive features such as the better-organized order of movement,
in addition to the energy constraint which poses no problem in the vehicular network,
because the nodes are connected to the vehicle’s electrical system, thus reducing a certain
number of known attacks in ad hoc networks aiming at consuming the energy sources of
the nodes. In addition, the time and position in VANET are known and are used for the
transmission of messages and also for their security. In addition, physical access to the
network is restricted by an authority to drivers and the service center, which can reduce
access to malicious entities [17].
Vehicular ad hoc networks have to deploy security requirements, including authentica-
tion, confidentiality, integrity, availability, non-repudiation, and privacy to cope with security
threats. These requirements use one or more security mechanisms designed to detect or pre-
vent a security attack [18–20].
60  Opportunistic Networks

4.3.2.1 Authentication
This requirement is one of the most critical system requirements. In VANET, it is crucial to
consider more information about the transmission node, as well as its identity, properties,
and geographical location. Therefore, the source node and the message broadcasting in
the network must be authenticated. The main purpose of authentication is to control the
authorization levels of the vehicle in the network. Two forms of authentication exist:

Entity authentication that helps to determine the type of equipment that is in communication.
It may be another vehicle, an RSU, or other equipment.
ID authentication that enables the node to define the transmitters of a given packet by
specifying a unique identifier for each vehicle. Using this technique prevents a vehicle
from claiming to have multiple identifiers and pretending to be more than one vehicle,
and thereby perpetrating an attack on the network.

Several attacks have been designed on authentication in VANETs, such as:

GPS spoofing: This consists of forcing the antenna, or module receiving the signal, to emit
a false GPS signal in order to change the location of the vehicle. That has an effect on
VANET routing, and the attacker can actually trick vehicles into believing they are at
another location.
The Sybil attack: This represents the case in which the malicious node is capable of claiming
multiple entities called Sybil nodes or false nodes. Thus, by posing as these different
identities, the malicious node can more easily compromise the general functioning of
the vehicle network.
Node impersonation: The attacker exploits the identity of another vehicle or RSU, phone,
or laptop in other to impersonate a legitimate node and take advantage of their
privileges.

4.3.2.2 Confidentiality
The objective of confidentiality is to make sure that the information is available only to
authorized entities. It is, therefore, the protection of data transmitted against passive attacks.
In order to provide this service, the mechanism used is cryptography, in particular encryp-
tion algorithms. Encryption can be achieved by symmetric or asymmetric cryptography. In
the case of symmetric cryptography, the stakeholders must share a secret key. This key is
used to encrypt and decrypt the message. One of the major problems with symmetric cryp-
tography is the creation of a secure channel for distributing the private key. In the instance
of asymmetric cryptography, each vehicle has a pair of keys: public and private keys. The
private key is known only by the source vehicle, while the public key is shared with all enti-
ties in the network.
Confidentiality can be vulnerable to many attacks, like:

Monitoring Attack: In this attack, the malicious tracks the entire network, listening to all
communication that occurs in vehicle-to-vehicle (V2V) mode or vehicle-to-infrastruc-
ture (V2I) mode. Once important data or information is received, the attacker forwards
it to the interested person.
Traffic Analysis Attack: This is the method by which network traffic is analyzed to deduce
data from communication patterns. The traffic information capturing from this attack
are vehicle ID, location of the user, the traveling route of the user, and many others.
 Secure Hierarchical Infrastructure  61

4.3.2.3 Integrity
This security requirement ensures that the transmitted messages are not willingly or
inadvertently altered. The message receiver checks the received message. It ensures that
the sender’s identifier remains the same throughout the transaction and that the message
received is the one that was sent. Integrity protects against destruction and alteration of the
message during transmission. If a corrupted message is accepted, it is considered that there
has been an integrity violation. To implement integrity, the system should prevent attacks
against message corruption, since the content of the message must always be reliable. Certain
security protocols use the electronic signature to make sure that the packet has not modified
during the transaction. Thus, upon arrival of the message, the signature is verified to judge
the integrity of the message.
Several attacks on integrity are distinct:

Bogus Information attack: The main goal of the Bogus Information attack is to forward
inaccurate or bogus information in the network for the attacker’s personal gain. The
attacker can be an intruder (outsider) or a legitimate user (insider). For instance, the
malicious node may send warning messages to the other nodes indicating that traffic
conditions are heavy to make its travel easier on the road.
Timing attack: This attack depends on estimating and analyzing the time required to con-
duct some process in safety applications to detect secret information. In VANETs, times
are a critical constraint. Thus, all data broadcasted between vehicles require arriving at
a specific time. However, when the intruder (malicious vehicles) gets a packet, he adds
an interval of time before sending it to the final destination, in order to create a delay.

4.3.2.4 Availability
The goal of availability is to guarantee permanent access to a service or resources. The net-
work and applications must remain available even if the network crashes. This requirement
not only protects the network, but it also allows it to withstand fault. Consequently, services
must remain available until the failure has been solved. To attain all receivers of messages, an
efficient routing protocol is required. Certain messages must remain contained at a definite
time, so as not to mislead the vehicles if the information is no longer relevant.
Several applications require a rapid response from the sensors or the ad hoc network
because the delay will make the message obsolete, which can cause dire situations. Accordingly,
it is essential that resources are available at the right time and place. Availability may face
different attacks, among them:

Denial of Service (DoS): It is the most common assault in traditional networks and
in VANETs. The main goal of the DoS attack is to render the services or resources
of the network unavailable for an indefinite time. This attack can be made through
injecting several useless data into the network (flooding attack). The scope of such a
kind of attack is generally broad, implying that this attack will involve several nodes.
Therefore, a DoS attack can spread on a huge geographical zone through several nodes
using multi-hop routing.
Black hole: The attacker in the black hole attack claims to possess an optimal path to the
destination and suggests routing the message through this route, and then cheats the
protocol for routing. This attack has the effect of destroying or misusing the packages
captured without transmitting them to the destination.
62  Opportunistic Networks

4.3.2.5 Non-repudiation
This requirement prevents an entity from denying having participated in communication. It
protects the system against the denial of a node, which indicates that it did not participate in
communication when it did. Non-repudiation allows the receiver to prove that he received
the message from a third party. Thus, for each message received, the sender can be clearly
identified. The non-repudiation guarantees that the evidence of an incident or action is
obtained, preserved, and made available to settle disagreements over whether or not an
action took place. Consequently, non-repudiation relies on authentication. This helps the
system to recognize the sender of the malicious packet. There are several threats in terms of
non-repudiation, such as:

Message Tampering: Attackers may alter messages transmitted in the V2V or V2I
communications during this attack. Message tampering attacks are induced by
authentication vulnerabilities.
Message Suppression/Fabrication: Attackers are able to deactivate themselves from
responding to any messages in two ways, either by disabling the connection between
vehicles or by modifying the application.

4.3.2.6 Privacy Preserving
In the vehicular network, the driver’s privacy is an essential aspect. Thus, the personal and
private information of a user should not be accessible by other users. However, the driver's
identity must not be inferred from the vehicle information (such as speed, geographical
location, the destination, etc.) propagated in the network. Through this information, the
position of the driver, vehicle movements, and tracking are more sensitive and must be taken
into account with care. Like authentication, privacy also faces the GPS spoofing attack and
many others such as:

Location tracking: The aim of the malicious node is to violate the privacy of a user by
trying to find confidential information about it. The attacker tracks the trajectory of the
target vehicle in order to get its position or route followed by that driver.
Tunneling: A GPS signal fails temporarily in tunnels, which leads to a temporary failure
to location information. The malicious node exploits this weak point and injects false
position information into the vehicle’s database before receiving the original position
update.

4.4 PRELIMINARIES

We present in this section the cryptographic algorithms that we use in our proposal. They are
the Elliptic Curve Diffie-Hellman (ECDH) Protocol and the Edwards-curve Digital Signature
(EdDSA) Algorithm.

4.4.1 Elliptic Curve Diffie-Hellman (ECDH)

Elliptic Curve Diffie-Hellman is a key agreement protocol that allows two entities to create a
secret key that will be used for private key operations. The public key created by each entity
is shared with the other. The methods on the elliptical curves are used to generate the secret
 Secure Hierarchical Infrastructure  63

key. We assume that Alice and Bob agree to a standard key cryptographic protocol for data
exchange. It is considered that they had no contact and that the mode of communication
available to them through the channel is only public. Both exchange public data or public
cryptographic keys. Each of them has a private key used to generate the shared key called
a public key. The two individuals agree on the same domain parameters. The steps of this
protocol are as follows [21]:

1. Alice and Bob choose a common elliptical curve E on a prime field Fp. They also have a
base point G € E (Fp) so that the subgroup generated by G has a greater group cardinal-
ity. This determines the strength of the method involved.
2. Alice chooses an integer a. It is a secret key that is not shared with anyone. This is
Alice’s private key. It then uses point multiplication and calculates the public key Ta =
a.G and sends Ta to Bob.
3. Bob also selects an integer b which becomes his private key, calculates Tb = b.G by
multiplying points and sends Tb to Alice.
4. Alice calculates a.Tb = a.b.G. This is done by multiplying points from Alice’s secret key
with Bob’s shared key.
5. Bob multiplies points between his private key and Alice’s public key and calculates
b.Ta = a.b.G. The only data that a spy can obtain concern the elliptical curve E, the
finite field Fp, and the points G, a.G, and b.G.

4.4.2 Edwards-curve Digital Signature (EdDSA)

The Edwards-curve Digital Signature Algorithm is a digital signature method based on the
Twisted Edwards curves using a Schnorr signature variant. A signature protocol published in
2011 by Bernstein et al. [22], EdDSA was originally intended to be used with the Curve25519
elliptical curve, but it may very well be used with any other elliptical curve. A signature con-
structed from this protocol is a couple (R, s) and is generated from:

E – the parameters defining the elliptical curve used.

P – a point on the high-order curve.
n – the order of point P.

The EdDSA key-pair consists of the following:

• The private key is generated from a random integer, named seed (which has to use a
similar bit length as a curve order). The seed is first hashed, then the last few bits, cor-
responding to the curve cofactor, are cleared; then the highest bit is cleared, and the
second-highest bit is set.
• The public key PubKey is a point on the elliptic curve, calculated by the EC point mul-
tiplication: PubKey = PrivKey * P (the private key, multiplied by the generator point P
for the curve).

Algorithms 4.1 and 4.2 respectively present the signature and the verification of signatures
performed by this protocol.
64  Opportunistic Networks

ALGORITHM 4.1  EDDSA SIGNATURE

Inputs: the private key, a; the associated public key, Q; the message, m
Outputs: (R, s) the signature associated with m
H(a, m) ← hash(a, m)
R ← H(a, m)P
H(R, Q, m) ← hash(R, Q, m)
s ← (H(a, m) + H(R, Q, m)a)[n]
return (R, s)

ALGORITHM 4.2  EDDSA VERIFICATION

Inputs: a message, m; the associated signature, (R, s); the public key associated with
the private key which served to sign the message, Q.
Outputs: returns true if the signature is correct, false otherwise
H(R, Q, m) ← hash(R, Q, m)
U ← 8sP
V ← 8R + 8H(R, Q, m)Q
if U ≠ V then
return False
end if
return True

We can show that the signature produced by Algorithm 4.1 will be validated by
Algorithm 4.2.
R  H  R,Q,m  Q  H  a,m  P  H  R,Q,m  aP
  H  a,m   H  R,Q,m  a  P
 sP

4.5 PROPOSED SCHEME

This section presents the description of the phases of our secure hierarchical infrastructure
scheme in the VANET.

4.5.1 Protocol Description
Our solution takes place in four phases:

• First is the main authentication phase between the different network entities which
allows VANET entities to authenticate with the CA in order to have a public key certifi-
cate for use in communication. This phase is executed by using a symmetrical approach
in order to secure the authentication packets exchanged. In addition, the EdDSA algo-
rithm for the generation of the public/private key pair, the generation of the signature
of a message, and its verification.
• The second phase is the authentication and communication phase between OBUs and
RSUs, where OBUs authenticate for a second time with RSUs for the purpose of having
two different keys. The first is a shared secret key for I2V or V2I communication, while
the other key is used for V2V communication.
 Secure Hierarchical Infrastructure  65

• The phase of message exchange between OBUs or V2V communication, using the KGi
group key issued by RSUs to encrypt the messages transmitted, and the use of the private
key to sign the hash of the message in order to ensure integrity and non-repudiation. In
addition, the use of a certificate makes it possible to guarantee the authentication and
the identity verification of the sender.
• The last phase is the revocation phase, of which the CA sends a list of revoked cer-
tificates to the OBUs. A certificate is revoked in the event that a vehicle declares the
theft or loss of its private key or in the event that an RSU suspects the behavior of a
vehicle. The vehicle receiving the list of revoked certificates can check the validity of a
sender’s certificate by searching for it in its list. In our solution, we proposed to share
the main list of revoked certificates to sub-lists according to the type (Professional,
Private, Personal, etc.) of the vehicle, which is a field in the certificate, in order to reduce
the response time. When a vehicle searches for a certificate whether it is revoked or not,
instead of searching the entire main list, it searches only in the list corresponding to the
type of vehicle.

The notations used in our proposal are illustrated in Table 4.1.

4.5.2 Registration Phase
In this work, we consider the same registration and authentication process to be used for
mutual authentication between CA→RSUs and CA→OBU. The vehicle to authenticate and
to have a public key certificate must execute a series of steps:

Table 4.1  The different notations used in the proposed solution.

Notation Description

CRSUi, COBUi, CCA RSUi cookie, OBUi cookie, CA cookie

KPOBUi The OBUi private key
KPUBOBUi The OBUi public key
Enk et Dek Encryption and decryption algorithms, respectively
Hash The hash calculated after using the hash function defined in SA
KOBUi The OBUi’s secret key
KCA TA’s secret key
IDrOBUi The real OBUi identifier
IDOBUi OBUi pseudo identifier
SigOBUi The digital signature of the OBUi
H The clock associated with the message to determine its freshness; it is the moment
when the message msg is sent
CertOBUi Vehicle certificate issued by CA
TypeOBUi The type of vehicle (Professional, Public, Personal, etc.)
KRSUi The RSUi public key
Texp The lifetime of the certificate.
IDCA The identifier of the certification authority
SigCA The certificate signature produced by the CA private key
KGi The group key of vehicles belonging to the same RSUi, used for V2V communications
KS The session key issued by the RSU
66  Opportunistic Networks

• Step 1: OBUi → CA: COBUi, SAOBUi.

The OBUi sends a message containing a cookie COBUi used to confirm that the OBUi
is communicating with the CA and SAOBUi. presented in the SA block that represents the
list of algorithms cryptographic supported by OBUi.
• Step 2: CA → OBUi: CCA, SATA, IDOBUi.
After the reception of the first message, the CA responds with a message similar to
the first message, which contains a CCA cookie with the same purpose as COBUi, SACA
as the type of cryptographic algorithm chosen and used in encrypted exchanges, and
IDOBUi as a session identifier.
• Step 3: OBUi → CA: CCA, Nonce, {IOBUi.P}, IDOBUi.
The OBUi receiving the message and checks the validity of the COBUi cookie. The
OBUi and CA agree together and publicly on an elliptical curve E (a, b, K), they choose
a finite field K in (Z/pZ) and an elliptical curve, and they choose together, and always
publicly, a point P located on the curve. Then the OBUi chooses an integer IOBUi and
sends a message that contains a header CCA, which is the same in the previous message,
the point of the elliptical curve IOBUiP and a nonce as a random for the design of the
keys, and IDOBUi as its session identifier.
• Step 4: CA → OBUi: COBUi, Nonce, {ICA.P}.
After reception of parameters and verification of the validity of CCA, the CA gener-
ates an integer KCA and returns a message which contains ICAP. At this point, the OBUi
and CA can calculate their secret keys KOBUi and KCA, respectively.

KOBUi  IOBUi  ICAP   ICA  IOBUiP   KCA   IOBUiICA  P

• Step 5: OBUi → CA: EnkKOBUi (COBUi, IDrOBUi, IDOBUi, Hash{S1, S2, S3, S4}).
After establishing the secret key, the OBUi sends a message containing a header
COBUi, a real identifier IDrOBUi, and the session identifier IDOBUi, plus a hash of the four
previous steps using the hash function defined in SA (SHA-256) and all encrypted with
the secret key (AES-128).
• Step 6: CA → OBUi: EnkKCA (IDOBUi, Hash{S1, S2, S3, S4}, KPUBCA).
Upon receiving this message and after decrypting it with the CA’s secret key, the CA
retrieves the IDrOBUi in order to generate a certificate and returns the message with a
hash of the four previous messages.
After checking the fingerprints (the hash) sent by the two entities, these fingerprints
must be the same in order to validate and confirm the agreement on the keys and the
algorithm to be negotiated, by carrying out an integrity check and authenticating the
messages traded. The CA public key is used to verify the signature of the latter in the
certificate issued to the OBUi.
• Step 7: The OBUi requests a certificate from CA in order to communicate with the other
entities in the network.
• Step 8: CA → OBUi: EnkKCA(KPUBOBUi, KPOBUi, SigOBUi, t, CertOBUi).
Upon receiving the message of the request, the CA checks the IDOBUi, executes the
EdDSA algorithm to generate the public/private key pair and the signature for the
OBUi, and sends a message encrypted with its secret key which contains the pair of
keys associated with the vehicle, the signature of the vehicle, and the certificate.
• The OBUi decrypts the message, extracts the key pair, the signature, and the certifi-
cate, and sends a Finished message to the CA indicating the end of the authentication
process.
• Upon receipt the previous message, CA sends back a Finished message to confirm the
end of the process.
 Secure Hierarchical Infrastructure  67

Figure 4.2 shows the different messages exchanged during the authentication phase.
The CA records in its database the information relating to the OBUi <IDOBUi, IDrOBUi,
KPOBUi, KPUBOBUi> to use in the event that a tracking has been launched against a vehicle in
bad behavior, for example, in order to obtain the true vehicle identity.
The format of a certificate:

CertOBUi : IDOBUi , KPUBOBUi , TypeOBUi , Texp , IDCA ,SigCA 

4.5.3 Authentication and Communication between OBUi and RSUi

A vehicle with a certificate issued by the CA can communicate with the other network
RSUs, and this can be done after mutual authentication between the OBUi and the RSUi
(Figure 4.3).

Figure 4.2  Registration phase.

68  Opportunistic Networks

Figure 4.3  Authentication and communication between OBUi and RSUi.

• Step 1: RSUi → OBUi: (IDRSUi, KPUBRSUi, CertRSUi)

The RSUi periodically broadcasts a message that includes its IDRSUi identity, its KRSUi
public key, and its CertRSUi certificate issued by the CA. This certificate allows the
OBUi to verify the validity of the RSUi public key. By intercepting this message, a OBUi
retrieves the public key from the RSUi and sends a series of authentication and secret
key generation messages following the ECDH algorithm.
• Step 2: OBUi → RSUi: EnkKPUBRSUi(COBUi, IOBUi.P, CertOBUi).
The OBUi sends a first message containing a COBUi cookie to fight against DoS attacks
and IOBUiP (point of the elliptical curve generated using ECDH algorithm); thus, its
CertOBUi certificate issued by the CA.
• Step 3: RSUi → OBUi: (COBUi, CRSUi, IRSUi.P, [IOBUi.P, IRSUi.P, IDOBUi]EnkKPRSUi)EnkPUBOBUi.
The RSUi when receiving the first message from OBUi, decrypts (COBUi, IOBUi.P,
CertOBUi)KPUBRSUi using its private key KPRSUi and checks the validity of the certificate
of OBUi CertOBUi. The RSUi then sends a message which contains its signature, the COBUi
vehicle cookie, and its CRSUi cookie, where the IDOBUi is a pseudo identifier used to pro-
tect user privacy.
• Step 4: OBUi → RSUi: (CRSUi, [IOBUi.P, IRSUi.P, IDRSUi]EnkKPOBUi)EnkKPUBRSUi.
The OBUi checks the signature [IOBUi.P, IRSUi.P, IDOBUi]KPRSUi using the public key
of the RSUi and checks the COBUi. If the cookies are different, it can directly interrupt
the communication because it perceives that it is exchanged with a malicious entity.
Otherwise, it continues the exchange by sending his signature [IOBUi.P, IRSUi.P, IDRSUi]
KPVi and the CRSUi. Similarly for the RSUi, when it receives the message, it first checks
the CRSUi cookie if it is valid, and it decrypts the signature.

At this time, the OBUi and the RSUi can calculate the symmetric and shared secret key KS,
used for communications between an OBUi and an RSUi. In addition to this key, the RSUi
generates another KGi key for vehicles after successful authentication. This key will allow
vehicles to communicate with each other in the same coverage area of the RSUi.
 Secure Hierarchical Infrastructure  69

The RSUi communicates with neighboring RSUs to obtain their group keys (V2V commu-
nication key). At the same time as the delivery of the KGi key to the vehicles, it sends the keys
of the neighboring RSUs obtained as well as their identifiers (a set of keys). Consequently, the
OBUi will have a list of group keys from neighboring RSUs in addition to that of its RSUi.
When receiving a message from an OBUj, the OBUi retrieves the identifier of the RSUj in
order to determine the corresponding key for decryption.
The use of two different keys for both I2V and V2V communication ensures confidentiality
especially at the level of I2V communication, because the use of a group key causes certain
drawbacks. For example, an OBUi requesting service with an RSUi encrypts its request with
the group key, and in the case that an OBUj receives this information, it can decrypt the
message and know its content, which violates the confidentiality of the message. However,
with the use of the secret key, the OBUi can send a request, and only the RSU can decrypt it
using its key shared with the OBUi, thus preserving the concept of confidentiality.

4.5.4 Communication between Vehicles

To send messages to other vehicles, the OBUi encrypts the message with the KGi key delivered
by its RSUi, signs the message content using the EdDSA algorithm with its private key, and
sends its certificate with the adding of a parking meter (Clock) for the message freshness
(Figure 4.4).

OBU i  OBU j  (IDOBUi , CertOBUi , Data  Msg, h  , SigOBUi  Hash  Data   Enk KGi , IDRSUi

When receiving a message, the OBUj checks the RSUi identifier, retrieves the corresponding
key and decrypts the message with the KGi key corresponding to the RSUi identifier. It verifies
the sender’s signature using the EdDSA algorithm and the time interval between the current
moment and the moment when the OBUi sent the message. This interval must not exceed a
predefined threshold (e.g. 300 ms). In addition, it checks the validity of the certificate for the
current date. If the certificate is not valid, it ignores the message; otherwise, the message is
accepted. The sender’s current position can be added to the sent message in order to avoid
replay attacks, where message receivers including the RSU can compare between the sender’s
current position and the position to which he sent the message in order to deduce whether
this entity is malicious or not.

4.5.5 Revocation
A vehicle whose behavior is detected to be abnormal – for example, misconduct or the send-
ing of false safety messages, etc. – is marked malicious. The RSUs are responsible for this
task and inform the CA to die whether or not their certificate is revoked. In addition to its
behavior, a vehicle may lose its private key, or its key may be stolen. So in these cases, the
vehicle certificate must be revoked and registered in the Certificate Revocation List (CRL)
revoked certificate list in order to inform the RSUs and vehicles belonging to the VANET.
This is illustrated in Figure 4.5.

Figure 4.4  Communication between vehicles.

70  Opportunistic Networks

Figure 4.5  The different steps of the revocation phase.

. The CA sends the revoked certificate for all RSUs.

1
2. The RSUs send the revoked certificate to the OBUi.
3. The OBUi updates its CRL following the previous request.

In this phase, we are proposing a new strategy for managing CRLs, in order to improve the
response time to search for the validity of a certificate. We will proceed by dividing the main
CRL into several sub-CRLs based on the type cited in the public key certificate; for example,
one under CRL for professional vehicles, one under CRL for private vehicles, etc.
When updating these sub-CRLs, the revoked certificate will register in a sub-list according
to its type, and when an OBUi wants to verify the certificate, it will not need to browse the
entire main list, it searches in the sub-list corresponding to the type of OBUj certificate.

4.6 SECURITY ANALYSIS

In this section, we analyze the security analysis and the formal verification to illustrate that
our model can reach the security objectives and requirements.

4.6.1 Analysis of Security Requirements

• Mutual authentication between the CA and the OBUs and between the CA and the
RSUs: It is ensured by the authentication process where the OBU authenticates with the
CA (and the CA authenticates with the OBU) and ensures that the latter is the entity it
claims by using cookies, in order to issue a public key certificate for the OBU.
 Secure Hierarchical Infrastructure  71

• Mutual authentication between RSUs and OBUs: It is carried out in the second phase of
the solution, where OBUs and RSUs exchange a series of messages in order to authen-
ticate each other and obtain a secret key. This secret sharing is carried out using the
ECDH algorithm.
In addition to this key, the RSU generates a group key KGi for all the OBUs in its
zone, which is used for V2V communication.
• Negotiating and choosing a cryptographic policy: It allows an algorithm to be easily
excluded in the event of vulnerability, in order to replace the cryptographic policy used
by a new policy supported by the two entities. To remedy the vulnerability of an algo-
rithm, we can define a duration during which a user can authenticate, and in case the
defined threshold is exceeded, we will change the cryptographic policy used.
• Confidentiality of authentication packets: Our solution ensures the confidentiality of
authentication packets, by encrypting them using the secret key shared between the
OBUs and CA.
• Protection against replay attacks: The use of nonce as random events in order to ward
off replay attacks, where the malicious replaying entity cannot calculate shared secrets.
• Protection against denial of service (DoS) attacks: The use of cookies allows our solu-
tions to remedy DoS attacks. Malicious OBUs generally start by spoofing the addresses
of other OBUs, then massively sending authentication requests in order to exhaust CA
resources, or using the latter as an amplifier or attack relay to OBUs from which they
initially usurped their address. Therefore, the exchange of cookies during the authen-
tication phase is essential, so that the CA server expends its resources only if the OBU
returns the CA cookie, to confirm that it is in use. This limits exchange with the alleged
entity, which limits this type of attack.
• Confidentiality of the exchanged packets: This is ensured by encrypting the exchanges
with the secret key KS in V2I or I2V communications and with the group key KGi in
V2V communications.
• Non-repudiation: The signature makes it possible to verify the identity of the sender of
this fact. Each message exchanged in the VANET must be signed with the private key
of the sender. This key is generated by the CA and OBUs after successful authentication,
and the signature is verified through the sender’s public key and the use of the EdDSA
algorithm. Signing allows the receiver of a message to authenticate the identity of the
sender and to ensure non-repudiation.
• Privacy preservation: The use of pseudo identifiers or real vehicle identifiers allows
users to preserve their private lives by exchanging messages anonymously. These pseudo
identifiers are issued by the CA during the authentication process.
• Availability: DoS attacks aim to exhaust the resources of a server and make the server
inaccessible. The use of cookies makes it possible to counter these types of attacks and
therefore to ensure server availability.
• Integrity: The hash functions ensure data integrity. In our solution, the hash of the
four messages sent in Steps 5 and 6 makes it possible to verify the integrity of these
exchanges.

4.6.2 Formal Verification
We used the AVISPA (Automatic Validation of Internet Security Protocols and Applications)
[23] tool to provide a formal modular and expressive language to specify the protocols and
their security properties. It integrates a different back end which uses a variety of automatic
machine analysis techniques. The AVISPA tool uses the Dolev-Yao intrusion model, where the
attacker can spy on all transmitted messages, usurp the identity of a legitimate entity (attack
by identity theft), and modify or inject messages, but he considers that the cryptography is
72  Opportunistic Networks

Figure 4.6  AVISPA architecture.

perfect, i.e. the attacker cannot break the cryptography. The AVISPA framework is shown in
Figure 4.6. The first step in using the tool is to present the analyzed protocol in a special lan-
guage called HLPSL (High Level Protocol Specification Language). The HLPSL specification
of the protocol is translated into the lower-level language called IF (Intermediate Format).
This translation is done by a translator called HLPSL2IF. This step is completely transparent
to the user.
The IF specification of the protocol is used as an input to the four different back ends:
On-the-fly Model-Checker (OFMC), CL-based Attack Searcher (CL-AtSe), SAT-based
Model-Checker (SATMC) and Tree-Automata-based Protocol Analyzer (TA4SP). These back
ends perform the analysis and formatting of the output that contains the results. These results
indicate whether there are problems in the protocol or not.
The primary goal of our proposed scheme is to verify that it can provide a reliable key
exchange between the entities of the VANET in order to secure the registration, authentica-
tion, and data transfer phases by using back-end servers. In Figures 4.7 through 4.10, we
present respectively the roles of vehicle, CA, RSU, and the authentication goals.
After running this specification with OFMC and CLAtSe back ends, we can conclude that
the proposed scheme can accomplish our goal and can resist those malicious attacks, such as
replay attacks, secrecy attacks, and DoS attacks, under the test of AVISPA. The outputs of the
model checking results are shown in Figures 4.11 and 4.12.

4.7 PERFORMANCE EVALUATION

This section evaluates the performances of our authentication protocol with other existing
protocols according to security and computation overhead.
 Secure Hierarchical Infrastructure  73

role vi (Vi, CA, RSU : agent, K: symmetric_key,

H : hash_func,
SND, RCV: channel (dy))
played_by Vi
def=
local
State: nat,
Cv, SAv, Cca, SAca, IDv, IDrv, Nonce, Certv, Sigv, KPv, KPca, KPrsu,
Crsu, IDrsu, Certrsu: text,
Finish : message,
KPUBv, KPUBca, KPUBrsu: public_key
const
r1, r2, r3 : protocol_id

init State := 1
transition
1. State = 1
/\SND (start) =|>
State' := 2
/\Cv' := new()
.
.
2. State = 2
/\SND (Cca', Nonce', IDv') =|>
State' := 3
/\Cv' := new()
/\RCV (Cv', Nonce')
3. State = 3
/\SND ((Cv'. IDv'. IDrv'. H(Cv', SAv',Cca', SAca', IDv',Cca',
Nonce', IDv', Cv', Nonce')). K')
=|>
.
.
State' := 5
/\SND(Finish)
/\RCV(Finish)
5. State = 5
/\RCV (IDrsu', KPUBrsu', Certrsu') =|>
.
.
/\SND ((Crsu'. IDrsu').KPv')
end role

Figure 4.7  Role of vehicle.

4.7.1 Comparison of Security Performance

We have compared the security performance of existing authentication protocols with our
protocol. As shown in Table 4.2, the proposal can provide the most comprehensive security
performance and check the security level.

4.7.2 Computation Overhead
This part evaluates the computation cost required by related protocols and our scheme. In
this context we used the rational arithmetic C/C++ library (MIRACL) [24] installed on a
computer with a 3.2G HZ CPU and 8G of memory in order to calculate the execution time
of such single operations used in existing and proposed protocols [25]. Table 4.3 illustrates
the operations and their computation overheads.
The overheads are the sum of the time consumed on both the vehicle side and the RSU
side. Table 4.4 indicates the operation numbers of the computation overheads for the existing
74  Opportunistic Networks

role ca (CA, Vi, RSU: agent, K: symmetric_key,

H : hash_func,
SND, RCV: channel (dy))
played_by CA
def=
local
State: nat,
Cv, SAv, Cca, SAca, IDv, IDrv, Nonce, Certv, Sigv, KPv, KPca: text,
Finish : message,
KPUBv, KPUBca: public_key
const
r1, r2, r3 : protocol_id

init State := 0
transition
1. State = 0
/\RCV (start) =|>
State' := 1
.
.
2. State = 1
/\RCV (Cca', Nonce', IDv') =|>
State' := 2
/\Cv' := new()
/\SND (Cv', Nonce')
3. State = 2
.
.
/\Certv' := new()
/\SND ((IDv'. KPUBca'. H(Cv', SAv',Cca', SAca', IDv',Cca', Nonce',
IDv', Cv', Nonce')). K')
4. State = 3
.
.
/\SND(Finish)
end role

Figure 4.8  Role of CA.

role rsu (Vi, CA, RSU : agent, K: symmetric_key,

H : hash_func,
SND, RCV: channel (dy))
played_by Vi
def=
local
State: nat,
Cv, IDv, Certv, KPv, KPrsu, Crsu, IDrsu, Certrsu: text,
KPUBv, KPUBrsu: public_key
const
r1, r2, r3 : protocol_id

init State := 5
transition
1. State = 1
.
.
/\RCV ((Crsu'. IDrsu').KPv')
end role

Figure 4.9  Role of RSU.

 Secure Hierarchical Infrastructure  75

goal

authentication_on r1
authentication_on r2
authentication_on r3

end goal

Figure 4.10  Analysis goals of our scheme.

Figure 4.11  Results reported by the OFMC back end.

protocols [12–14] and the proposed protocol. We demonstrate the computation cost for
OBU and RSU in Figure 4.13. It is observed that the proposed protocol is faster than the
protocols adopted in [12–14]; this is due to the choice of lightweight operations to make the
mutual authentication.

4.8 CONCLUSION

The security of vehicular networks is a prerequisite for their deployment. Given that the fact
that these networks are a category of wireless networks and the importance of the informa-
tion exchanged which can endanger the lives of users, the security of these networks does
not stop to receive particular interest from research communities in academic and industrial
circles. Among the problems that have arisen in these networks is the problem of privacy
preservation of users. Our solution allows vehicles (OBUs) and RSUs to authenticate with
the certification authority so that they are legal entities in the network. This authentication
faces attacks such as denial of service attacks and replay attacks using cookies and nonce,
respectively. In addition, double authentication between OBUs and RSUs makes it possible
to minimize access to malicious entities in the network. Also, the use of a symmetric key
during V2I and V2V communication makes it possible to reduce the calculation time and
the verification time compared to other protocols using the asymmetric approach. The use
76  Opportunistic Networks

Figure 4.12  Results reported by the CL-AtSe back end.

Figure 4.13  Computation cost of different schemes.

of the EdDSA algorithm makes it possible to speed up the execution of the authentication
process, especially at the level of key management, message signature, and verification of
this signature. On the other hand, the creation of sub-lists of revoked certificates based on
vehicle type makes it possible to minimize the response time by looking for a certificate
whether it is revoked or not. In summary, our solution ensures basic security requirements
 Secure Hierarchical Infrastructure  77

Table 4.2  Security performance comparison.

Security features [10] [11] [12] [13] [14] Proposed
Scheme

Mutual authentication No Yes Yes Yes Yes Yes

Provides message No No Yes Yes Yes Yes
confidentiality
Provides message No No Yes Yes Yes Yes
integrity
Providing non-repudiation No No Yes No No Yes
Privacy preservation No Yes No Yes Yes Yes
Resistance against the No No No No No Yes
DoS attack
Resistance against the No Yes No Yes Yes Yes
replay attack

Table 4.3  Computation overhead of a single operation.

Operations Description Time (milliseconds)

PM Point Multiplication 2.258

BP Bilinear Pairing 6.443
H Hash (SHA-256) 0.021
EXP Exponentiation in Bilinear Group 3.212
ENC AES-128 Encryption 0.902
DEC AES-128 Decryption 7.357
MM Modular Multiplication 1.657
MP Modular Square Root 2.942
MTP Map-to-Point Hash Function 2.258
SING/VER Signature/Verification EdDSA 3.21

Table 4.4  Comparison of computation overheads.

Protocols Operation Numbers

Vehicle’s side RSU’s side Total (ms)

[12] PM + 2MTP + 5BP 2BP + 4MTP 48.352

[13] 4PM + 5H + 2EXP 3BP + 5H + 2EXP 41.419
[14] 7PM 7PM 30.95
Our 2ENC+DEC+SING/VER+2PM ENC+2DEC+SING/VER +PM 27.971
78  Opportunistic Networks

such as integrity, confidentiality, non-repudiation, and availability, as well as the preservation

of users’ privacy by using the real identifier.

REFERENCES

[1] Chavhan, S., Gupta, D., Chandana, B. N., Khanna, A., and Rodrigues, J. J. (2019). IoT-based
Context Aware Intelligent Public Transport System in a metropolitan area. IEEE Internet of
Things Journal, 7(7), 6023–6034.
[2] Zeadally, S., Hunt, R., Chen, Y. S., Irwin, A., and Hassan, A. (2012). Vehicular ad hoc networks
(VANETS): Status, results, and challenges. Telecommun. Syst., 50(4), 217–241.
[3] Eze, J., Zhang, S., Liu, E., and Eze, E. (2017, September). Cognitive radio technology assisted
vehicular ad-hoc networks (VANETs): Current status, challenges, and research trends. In 2017
23rd International Conference on Automation and Computing (ICAC) (pp. 1–6). IEEE.
[4] Sheikh, M. S., Liang, J., and Wang, W. (2019). A survey of security services, attacks, and applica-
tions for vehicular ad hoc networks (VANETs). Sensors, 19(16), 3589.
[5] Raya, M., Aziz, A., and Hubaux, J. P. (2006, September). Efficient secure aggregation in VANETs.
In Proceedings of the 3rd International Workshop on Vehicular Ad hoc Networks (pp. 67–75).
[6] Lin, X., Sun, X., Ho, P. H., and Shen, X. (2007). GSIS: A secure and privacy-preserving protocol
for vehicular communications. IEEE Transactions on Vehicular Technology, 56(6), 3442–3456.
[7] Bendouma, A. and Bensaber, B. A. (2017, May). RSU authentication by aggregation in VANET
using an interaction zone. In Proceedings of IEEE International Conference on Communications
(ICC) (pp. 1–6). IEEE.
[8] Vijayabharathi, V. and Malarchelvi, P. S. K. (2014, February). Implementing HMAC in expedite
message authentication protocol for VANET. In Proceedings of International Conference of
Information Communication and Embedded Systems (ICICES) (pp. 1–5). IEEE.
[9] Sakhreliya, S. C. and Pandya, N. H. (2014, December). PKI-SC: Public key infrastructure using
symmetric key cryptography for authentication in VANETs. In Proceedings of IEEE International
Conference Computational Intelligence and Computing Research (ICCIC) (pp. 1–6). IEEE.
[10] Das, A., Chowdary, D. R., and Rai, A. (2011). An efficient cross authentication protocol in
VANET hierarchical model. Int. J. Mob. Adhoc Netw., 1(1), 128–136.
[11] Li, J. S. and Liu, K. H. (2013). A lightweight identity authentication protocol for vehicular net-
works. Telecommunication systems, 53(4), 425–438.
[12] Han, M., Hua, L., and Ma, S. (2017). A self-authentication and deniable efficient group key
agreement protocol for VANET. KSII Transactions on Internet and Information Systems, 11(7),
3678–3698.
[13] He, D., Wang, D., Xie, Q., and Chen, K. (2017). Anonymous handover authentication protocol
for mobile wireless networks with conditional privacy preservation. Science China Information
Sciences, 60(5), 052104.
[14] Li, X., Han, Y., Gao, J., and Niu, J. (2019). Secure hierarchical authentication protocol in VANET.
IET Information Security, 14(1), 99–110.
[15] Goyal, A. K., Agarwal, G., and Tripathi, A. K. (2019). Network architectures, challenges, secu-
rity attacks, research domains and research methodologies in VANET: A survey. International
Journal of Computer Network and Information Security, 10(10), 37–44.
[16] Kaibalina, N. and Rizvi, A. E. M. (2018, October). Security and privacy in VANETs. In Proceedings
of IEEE 12th International Conference on Application of Information and Communication
Technologies (AICT). IEEE.
[17] Elsadig, M. A. and Fadlalla, Y. A. (2016). VANETs security issues and challenges: A survey.
Indian Journal of Science and Technology, 9(28), 1–8.
[18] Zhu, L., Zhang, Z., and Xu, C. (2017). Security and privacy preservation in VANET. In: Secure
and Privacy-Preserving Data Communication in Internet of Things. Springer Briefs in Electrical
and Computer Engineering, Springer, Singapore (pp. 53–76).
 Secure Hierarchical Infrastructure  79

[19] Al Junaid, M. A. H., Syed, A. A., Warip, M. N. M., Azir, K. N. F. K., and Romli, N. H. (2018).
Classification of security attacks in VANET: A review of requirements and perspectives. In
MATEC Web of Conferences, (Vol. 150, 06038). EDP Sciences.
[20] Hasrouny, H., Samhat, A. E., Bassil, C., and Laouiti, A. (2017). VANet security challenges and
solutions: A survey. Vehicular Communications, 7, 7–20.
[21] Fournaris, A. P., Zafeirakis, I., Koulamas, C., Sklavos, N., and Koufopavlou, O. (2015, May).
Designing efficient elliptic curve Diffie-Hellman accelerators for embedded systems. In IEEE
International Symposium on Circuits and Systems (ISCAS) (pp. 2025–2028), IEEE.
[22] Bernstein, D. J., Josefsson, S., Lange, T., Schwabe, P., and Yang, B. Y. (2015). EdDSA for more
curves. Cryptology ePrint Archive, 2015.
[23] AVISPA Project, http://www.avispa-project.org/
[24] MIRACL Library, http://www.shmus.ie/index.php.
[25] Wang, P., Liu, Y., and Lv, S. (2019). An improved lightweight identity authentication protocol for
VANET. Journal of Internet Technology, 20(5), 1491–1504.
Chapter 5

Simulation Tools for Opportunistic Networks

How to Set Up and Simulate the ONE Simulator
Md. Sharif Hossen
Comilla University, Cumilla, Bangladesh

CONTENTS

5.1 Introduction............................................................................................................... 81
5.2 Routing Protocols and Mobility Models.................................................................... 82
5.3 Simulation Tools .......................................................................................................84
5.3.1 ONE Simulator............................................................................................... 84
5.3.2 NS-2............................................................................................................... 84
5.3.3 OMNeT++...................................................................................................... 86
5.4 How to Install and Run the ONE Simulator.............................................................. 86
5.5 Default Settings in the ONE Simulator...................................................................... 89
5.6 Simulation Configuration for the ONE Simulator...................................................... 90
5.7 Simulated Results and Discussion with Analysis........................................................ 91
5.7.1 Delivery Probability........................................................................................ 91
5.7.2 Average Latency.............................................................................................. 91
5.7.3 Overhead Ratio............................................................................................... 93
5.7.4 Average Hop Count .......................................................................................96
5.7.5 Average Buffer Time........................................................................................ 96
5.8 Conclusion.................................................................................................................99
References........................................................................................................................ 100

5.1 INTRODUCTION

Delay/Disruption-Tolerant Networks (DTNs) [1], also called Intermittently Connected

Mobile Networks (ICMNs) [2], are mobile ad hoc networks where there is no consistent
path from one node to another, which is essentially required for successful communication.
These networks can be seen in the areas of satellite communication [3], tracking wildlife [4],
military applications, vehicular networks, etc. [5, 6].
Since the connection from one node to another node is not known in mobile networks, the
route is selected by choosing intermediate links, and hence it is a critical problem of DTN
communication. These intermediate links follow the “store-carry-and-forward” approach in
which nodes keep the information in their buffers and then carry them until finding a suit-
able node to forward [7]. Figure 5.1 illustrates the approach of the store-carry-and-forward
approach, where source A sends a message copy to the neighboring node B which stores
the copy into its buffer. Then, it carries the copy and forwards to the neighboring node C.
Repeating this process, node C reaches the destination D and then finally sends it to D.
Routing for DTNs is classified into two ways: forwarding- and replication-based. This
chapter represents an algorithmic and pictorial discussion and analysis of forwarding and

81
82  Opportunistic Networks

Figure 5.1  Store and forward strategy.

replication-based routing approaches, for example, first-contact, direct-delivery, Epidemic,

and PRoPHET.
The chapter has been outlined as follows. Section 5.2 illustrates the routing algorithms
with pictorial representation. Section 5.3 includes the discussion of simulation tools.
Section 5.4 gives details of how to install and run the ONE simulator. Section 5.5 illus-
trates the default settings in the ONE simulator. The simulation configuration of ONE is
mentioned in Section 5.6. Section 5.7 elaborates on the simulated results with performance
evaluation. Section 5.8 provides the concluding remarks.

5.2 ROUTING PROTOCOLS AND MOBILITY MODELS

This section illustrates the algorithmic and pictorial discussion of DTN routing protocols
which can be [8] either forwarding-based or replication-based. Forwarding routings do not
occupy many network resources since only a copy remains in the network at the network
lifetime [9,10]. Existing forwarding-based routings are direct-delivery [11] and first-contact
[9,12]. In the first-contact routing, a single message is forwarded to the first available contact
to meet the destination. As shown in Figure 5.2(a), at first say, source node S encounters node
C, then C forwards the message directly to D. In direct-delivery, S itself delivers message cop-
ies only to the final recipient D as shown in Figure 5.2(b).

Figure 5.2  Forwarding-based routing.

 Simulation Tools for Opportunistic Networks  83

ALGORITHM (FIRST-CONTACT)

A node can forward only a single message copy.

Step 1: if first available contacts are found,

then a node forwards a single message copy to that contact and deletes its
own copy form buffer,
else
a node keeps the copy in its own buffer and finds next available contact.
Step 2: Continue Step 1 until the node meets the destination.

ALGORITHM (DIRECT-DELIVERY)

Step 1: if Source node directly finds the destination node,

then Source forwards the copy to it and stops the communication.
else
source node moves and waits for the destination.
Step 2: Continue Step 1 until the node meets the destination.

ALGORITHM (EPIDEMIC)

Step 1: if there are no message copies in common,

Each node can forward message copies to its neighboring nodes.
else
no copies can be forwarded.
Step 2: Continue Step 1 until the node meets the destination.

ALGORITHM (PROPHET)

Each node can forward a message copy to another.

Step 1: if a connection’s state changes and a new connection is created to a node, then
the node finds its peer and collects all essages from this peer, and finds new
copies of message that this routing does not hold.
Step 2: if a node has not already traversed another then,
the node can forward a message copy to another.
else
no message copies can be sent.
Step 3: Continue Step 1 and 2 until the node meets the destination.
84  Opportunistic Networks

Figure 5.3  Replication-based routing.

In replication-based routing, redundant copies of the same message remain in the network.
Examples of these routings are Epidemic [12], PRoPHET [13], MaxProp [14], RAPID [15],
Spray and Wait (SNW) [16], and Spray and Focus (SNF) [17]. In this chapter, only Epidemic
and PRoPHET have been used for investigation.
Epidemic routing consumes many network resources because of its flooding nature, where
a node can send more than one copy to its neighbors which do not have the same message.
In Figure 5.3(a), S sends copies to its neighbor nodes; these nodes then forward copies to the
next nodes assuming no copy in common to the nodes.
In PRoPHET routing, depending on the higher value of the probability of a node to
contact the destination, the source can forward a message to a suitable node as shown in
Figure 5.3(b).
In this investigation, the Shortest Path Map-Based Movement (SPMBM) mobility model
has been used that uses a map-described environment to restrict node movement. This model
uses the Dijkstra algorithm to find the shortest route to reach the destination. Although all
the points on a map have equal probability, depending on users’ movements or preferences,
map data may contain some points of interest (POI) groups gathering people in different
places, e.g., tourist attractions and supermarkets. This model shows good performance com-
pared to Random Walk and Random Direction [18].

5.3 SIMULATION TOOLS

In this section, three types of simulation tools are briefly discussed.

5.3.1 ONE Simulator
Opportunistic Network Environment (ONE) is an event-driven Java-based simulator which
is particularly used for analyzing DTN routings [11]. Real-time nodes’ traces can be seen
using a graphical user interface. Reports are generated in a text file. All the settings can be set
using different file extensions internally and externally. Hence, the version of 1.5.1 RC2 of the
simulator has been used [19]. Figure 5.4 shows the simulation scenario of the ONE simulator.

5.3.2 NS-2
A network simulator called NS-2 along with Linux is used as a simulation tool. It has an
object tool command language shell (OTCL) as the user interface, which allows scripts writ-
ten in TCL format to be executed. It is a discrete event simulator along with the object-
oriented concepts and is used mostly by the ad hoc networking community and researchers
to determine the performance of routings. It uses two languages: (1) C++, which runs in the
back end, and (2) tool command language (TCL), which runs in the front end. Moreover, the
components of NS-2 include NS, NAM (network animator), pre-processing, and post-anal-
ysis for the simulator, visualization of NS output, handwritten TCL, and trace analysis using
AWK scripts, respectively. Figure 5.5 shows the scenario of black hole simulation [20, 21].
 Simulation Tools for Opportunistic Networks  85

Figure 5.4  Simulation scenario of ONE.

Figure 5.5  The NAM scenario for black hole simulation.

86  Opportunistic Networks

Figure 5.6  Simulation scenario of OMNeT++.

5.3.3 OMNeT++
OMNeT++ is an event-driven framework to build and implement the emulated scenario of
mobile ad hoc networks. It is developed using the C++ language. Figure 5.6 shows the simu-
lation scenario of OMNeT++.

5.4 HOW TO INSTALL AND RUN THE ONE SIMULATOR

This section will discuss how to install and run the ONE simulator. The simplest way to
install the ONE simulator is to use Eclipse.
The following steps are needed to install the ONE simulator.

i . Download and install Java JDK

ii. Download ONE simulator
iii. Download and extract Eclipse and
iv. Integrate the ONE simulator in Eclipse

These four steps are discussed briefly with screenshots.

i. Download and install Java JDK

First, go to the following link or search on Google by typing “Java JDK”:
https://www.oracle.com/java/technologies/javase-downloads.html
Then, download and install the JDK version as your computer’s configuration.
ii. Download ONE simulator
Download the ONE simulator from any of the following two links.
http://akeranen.github.io/the-one/
https://www.netlab.tkk.fi/tutkimus/dtn/theone/
 Simulation Tools for Opportunistic Networks  87

The file that you will download is a zip file. Consider naming the file “ONE.zip”.
As of this writing, the latest version release of ONE is 1.6.0. After extracting the file,
a folder named “ONE” will be created that will be used later during the integration of
the ONE simulator in Eclipse.
iii. Download and extract Eclipse
Go to the following link or search on Google by typing “eclipse for java developers”.
https://www.eclipse.org/downloads/packages/release/kepler/sr1/eclipse-ide-java-
developers
You can download the eclipse according to your PC’s operating systems.
iv. Integrate the ONE simulator in Eclipse
The step-by-step integration of the ONE simulator in Eclipse is illustrated as follows:
First, create a Java project named “ONESim” in Eclipse as follows.
File → New → Java Project
This means that you click on the “File” tab, then click “New”, and then “Java
Project”.
After clicking “Java Project”, enter the project name as “ONESim”. Click “Next”
and then the “Finish” button. A Java project named “ONESim” will be created.
Now, ONE simulation will be integrated into Eclipse. Open the ONE folder that you
have extracted. Copy the “src” folder in the extracted file. Paste and replace the “src”
folder in the “ONESim” Java project.
It needs to add some jar files. Jar files are included in the “lib” subfolder of the “src”
folder. To integrate the jar files, right-click on the “ONESim” project, then go to the
following director (Figure 5.7).
Build Path → Configure Build Path

Then, a window will appear, and there, click on “ModulePath”. After that, click on the
button “Add JARs” on the right side and then a “JAR Selection” window will appear. Here,
it is necessary to select the jar files under the “lib” subfolder of the “src” folder. Then, click
the “OK” button. Finally, click the “Apply and Close” button. It is now integrated the ONE
simulator completely. It is shown in Figure 5.8. It is necessary to download a “junit” jar file
that will not be found into the “lib” folder. The link is as follows:

https://repo1.maven.org/maven2/junit/junit/4.12-beta-3/junit-4.12-beta-3.jar

To run the ONE simulator, right-click on the java project and choose “Run As” and then
“Run Configurations”.
Hence, a window will appear as shown in Figure 5.9. Here, select the main class as
“DTNSim” in the Main tab. Then, go to the Arguments tab. Select the “Other” button under
the “Working director” and click on the “Workspace” button. Finally, select the “src” folder,

Figure 5.7  Directory of “Configure Build Path”.

88  Opportunistic Networks

Figure 5.8  Adding jar files.

Figure 5.9  Executing the “ONESim” project.

 Simulation Tools for Opportunistic Networks  89

click the “OK” button, and then the “Run” button. Now, the simulation scenario of ONE
simulator, shown in Figure 5.4, will appear. This process of installing and running ONE simu-
lator is shown at https://www.youtube.com/watch?v=pH8MKr0pjeE.

5.5 DEFAULT SETTINGS IN THE ONE SIMULATOR

There is a default setting file (“defaultsettings.txt”) for the simulation where user can set
up the configuration of ONE simulation settings. Figure 5.10 shows scenario settings where
scenario name, update interval, and simulation time are given.
Nodes in the network can form a group which requires an interface to communicate within
a range with different transmit speeds. Two types of interfaces, Bluetooth and High speed,
can be used as a simple broadcast interface. The number of host groups can be described as
“Scenario.nrofHostGroups=6”.
If the user does not provide any information, a common setting for all groups will be
applied. Here, a common group does not contain any group name; just have only “Group”,
as shown in Figure 5.11.
For example, the first three groups shown in Figure 5.12 are not assigned by the number of
nodes, buffer space, interface, TTL, so they use the aforementioned common settings shown
in Figure 5.11.

Figure 5.10  Scenario settings.

Figure 5.11  Common setting for all groups.

Figure 5.12  First three groups in “defaultsettings.txt” file.

90  Opportunistic Networks

The remaining three groups consist of two nodes. Hence, the total number of nodes is
(40 × 3) + (2 × 3) = 126. In the message creation parameters as given below, this information
is needed in “Events1.host=0,125”, as shown in Figure 5.13. Message generation rates are
determined by updating the events’ interval.
The number of reports and reports’ names with the directory of the generated reports can
be given as shown in Figure 5.14.

5.6 SIMULATION CONFIGURATION FOR THE ONE SIMULATOR

Simulation settings are specified in Table 5.1. Four parameters have been considered fixed
for any change.

Message generation rate = 2 (msg/min)

Number of nodes = 80
Buffer size = 5 MB
Message TTL = 300 minutes

For example, for varying message generation rates 2, 3, 4, 5, and 6, three parameters are
considered constant as mentioned earlier (number of nodes = 80, buffer size = 5 MB, message

Figure 5.13  Event generations in “defaultsettings.txt” file.

Figure 5.14  Report settings.

 Simulation Tools for Opportunistic Networks  91

Table 5.1  Simulation settings.

Parameters Value

Simulation duration 12 hr
Message copies 2, 3, 4, 5, 6 (msg/min)
Nodes 20, 50, 80, 100, 120
Buffer size 5, 10, 15, 20, 30 (MB)
Message TTL 60, 120, 180, 240, 300 (min)
Movement model SPMBM
Message size 500 KB to 1 MB
Forwarding-based routing First-contact, direct-delivery
Replication-based routing Epidemic, PRoPHET
Area 4500 × 3400 m

TTL = 300 minutes). Two types of groups have been considered, pedestrians (P) with speed
of 0.5–1.5 m/s, and cars (C) with speed of 2.7–13.9 m/s. Bluetooth with speed 250 kbps and
range 10 m has been used as an interface.

5.7 SIMULATED RESULTS AND DISCUSSION WITH ANALYSIS

In this section, the performance analysis of the aforementioned routings is discussed. Hence,
the analysis is done on five performance metrics, namely delivery probability, average latency,
overhead ratio, average hop count, and average buffer time. These performance metrics are
obtained with varying four parameters: message generation rates, number of nodes, buffer
size, and time-to-live.
After completing the simulation, the calculated delivery probability, average latency,
overhead ratio, average hop count, and average buffer time are stored in the file
“MessageStatsReport.txt”. This file can be found in the following directory of the project
“ONESim” of your computer’s drive.

ONESim → src → reports → MessageStatsReport.txt

Figure 5.15 shows the generated message stats reports.

5.7.1 Delivery Probability
Delivery probability is calculated by dividing the number of messages received by the destina-
tion and the messages sent from the source. With increasing message copies, mobile nodes,
and time-to-live, the direct-delivery routing shows better performance, Epidemic indicates
lower results, as shown in Figures 5.16, 5.17, and 5.19, respectively. However, for varying
the buffer sizes, first-contact exhibits good delivery, while Epidemic shows lower delivery as
shown in Figure 5.18. In all cases, Epidemic represents lower delivery.

5.7.2 Average Latency
Average latency is calculated from the difference of time between messages produced by
the source and the messages got to the destination. With the increase of message copies,
direct-delivery routing shows lower latency when the number of message copies is 2, while
Epidemic routing shows lower latency when the message copies are greater than 2, as shown
92  Opportunistic Networks

Figure 5.15  Message stats reports.

Figure 5.16  Delivery probability vs. message generation rate.

Figure 5.17  Delivery probability vs. mobile nodes.

 Simulation Tools for Opportunistic Networks  93

Figure 5.18  Delivery probability vs. buffer size.

Figure 5.19  Delivery probability vs. TTL.

in Figure 5.20. Direct-delivery shows lower latency compared to the other considered rout-
ings when the mobile nodes are more than 50, whereas Epidemic indicates lower latency
when mobile nodes are no more than 50, as shown in Figure 5.21. With the increase of buffer
sizes, i.e., 5, 10, 15, 20, and 30 MB, direct-delivery routing shows lower average latency while
first-contact shows higher latency, as shown in Figure 5.22. Direct-delivery shows the lowest
latency when TTL = 6 minutes, while Epidemic indicates lower latency when TTL is less than
6 minutes, as shown in Figure 5.23.

5.7.3 Overhead Ratio
It determines how many copies of the same message are sent redundantly to reach a copy
successfully to the destination. In the case of overhead ratio, direct-delivery shows zero over-
head since the source node directly sends the message to the destination without sharing the
message to any node and hence exhibits better performance for varying message copies per
minute, the number of nodes, buffer size, and TTL as shown in Figures 5.24 through 5.27.
Whereas Epidemic routing shows the higher overhead ratio compared to direct-delivery, first-
contact, and PRoPHET routing protocols (Figures 5.24 to 5.27). Therefore, direct-delivery
shows good performance in terms of overhead ratio.
94  Opportunistic Networks

Figure 5.20  Average latency vs. message generation rate.

Figure 5.21  Average latency vs. mobile nodes.

Figure 5.22  Average latency vs. buffer size.

 Simulation Tools for Opportunistic Networks  95

Figure 5.23  Average latency vs. TTL.

Figure 5.24  Overhead ratio vs. message generation rate.

Figure 5.25  Overhead ratio vs. mobile nodes.

96  Opportunistic Networks

Figure 5.26  Overhead ratio vs. buffer size.

Figure 5.27  Overhead ratio vs. TTL.

5.7.4 Average Hop Count

Average hop count indicates the average number of intermediate nodes to build a successful
connection. In the case of average hop count, direct-delivery shows the lowest value, and its
value is 1 for varying all cases, i.e., message copies, mobile nodes, buffer size, and TTL as
shown in Figures 5.28 to 5.31. First-contact routing shows high overhead ratio among the
considered routing techniques (Figures 5.28 to 5.31).

5.7.5 Average Buffer Time

For varying message copies shown in Figure 5.32, the average buffer time of direct-delivery
is decreasing but shows value among all routings. With the increase of mobile nodes, the buf-
fer time of direct-delivery is increasing and takes a higher time than all routings as shown in
Figure 5.33. As shown in Figure 5.34, when buffer size is greater than 10 MB, direct-delivery
shows higher but constant (approximately 18000s) buffer time. With the increase of TTL, the
buffer time of direct-delivery is increasing, as shown in Figure 5.35. In all cases, as shown in
Figures 5.32 to 5.35, direct-delivery represents a higher average buffer time.
 Simulation Tools for Opportunistic Networks  97

Figure 5.28  Average hop count vs. message generation rate.

Figure 5.29  Average hop count vs. mobile nodes.

Figure 5.30  Average hop count vs. buffer size.

98  Opportunistic Networks

Figure 5.31  Average hop count vs. TTL.

Figure 5.32  Average buffer time vs. message generation rate.

Figure 5.33  Average buffer time vs. mobile nodes.

 Simulation Tools for Opportunistic Networks  99

Figure 5.34  Average buffer time vs. buffer size.

Figure 5.35  Average buffer time vs. TTL.

5.8 CONCLUSION

Delay tolerant network is a kind of ad hoc network having many applications in real life
like space communications, wildlife tracking, satellite communication, and so on, where the
establishment of an infrastructure-based network is often impossible. Due to the mobility
pattern and nodes’ behavior of randomly intermittent path selection, the routing in such
applications is difficult and challenging. So, the importance of simulating such a scenario is
needed where a lot of simulation tools such as NS-2, OMNeT++, ONE, JSim, and OPNET
are most usable. This chapter has included the discussion of how to install and integrate the
ONE simulator, how to set up the simulation settings in ONE, and how to compare the rout-
ing performance using ONE. Hence, the performance of DTN routings, e.g., first-contact,
direct-delivery, Epidemic, and PRoPHET has been evaluated by varying the message genera-
tion rate, the number of mobile nodes, buffer size, and TTL in terms of delivery ratio, average
latency, overhead, average hop count, and buffer time. Hence, direct-delivery routing shows
satisfactory performance in terms of the considered metrics.
100  Opportunistic Networks

ACKNOWLEDGEMENTS

The author would like to thank the University Grant Commission of Bangladesh for p
­ roviding
funding at Comilla University for this research.

REFERENCES

[1] M.S. Hossen, “DTN Routing Protocols on Two Distinct Geographical Regions in an Opportunistic
Network: An Analysis,” Journal of Wireless Personal Communications, 108(2), pp. 839–851,
2019.
[2] M.I. Talukdar and M.S. Hossen, “Performance Analysis of DTN Routing Protocols: Single-Copy
and Multi-Copy in ICMN Scenario,” In Proceedings of International Conference on Innovations
in Science, Engineering and Technology (ICISET-2018), IEEE, International Islamic University
Chittagong (IIUC), Bangladesh, October 2018.
[3] M.S. Hossen and M.S. Rahim, “On the Optimization of Number of Message Copies for Multi-
Copy Routing Protocols in Scalable Delay-Tolerant Networks,” In Proceedings of International
Conference on Computer, Communication, Chemical, Materials and Electronic Engineering
(IC4ME2), pp. 163–167, 2016.
[4] P. Juang, H. Oki, Y. Wang, M. Martonosi, L.S. Peh and D. Rubenstein, “Energy-Efficient
Computing for Wildlife Tracking: Design Tradeoffs and Early Experiences with Zebranet,” In
Proceedings of ACM ASPLOS, San Jose, CA, December 2002, pp. 96–107.
[5] M.I. Talukdar and M.S. Hossen, “Selecting Mobility Model and Routing Protocol for Establishing
Emergency Communication in a Congested City for Delay-Tolerant Network,” International
Journal of Sensor Networks and Data Communications, 8(1), pp. 1–9, 2019.
[6] T. Spyropoulos, K. Psounis and C.S. Raghavendra, “Efficient Routing in Intermittently Connected
Mobile Networks: the Single-Copy Case,” In IEEE/ACM Transactions on Networking, pp. 63–
76, February 2008.
[7] M.S. Hossen and M.S. Rahim, “Analysis of Delay-Tolerant Routing Protocols Using the Impact
of Mobility Models,” Journal of Scalable Computing: Practice and Experience, 20(1), pp. 39–48,
2019.
[8] M.S. Hossen, M.M. Billah and S. Yasmin, “Impact of Buffer Size and TTL on DTN Routing
Protocols in Intermittently Connected Mobile Networks,” International Journal of Engineering
and Technology, 7(3), pp. 1735–1739, 2018.
[9] M.S. Islam, S.I. Thaky and M.S. Hossen, “Performance Evaluation of Delay-Tolerant Routing
Protocols on Bangladesh Map,” International Conference on Advanced Computing and
Intelligent Engineering (ICACIE-2018), Springer, 2018.
[10] M.S. Hossen, M.T. Ahmed and M.S. Rahim, “Effects of Buffer Size and Mobility Models on
the Optimization of Number of Message Copies for Multi-Copy Routing Protocols in Scalable
Delay-Tolerant Networks,” International Conference on Innovations in Science, Engineering
and Technology (ICISET-2016), IEEE, October 2016.
[11] A. Keränen, J. Ott, and T. Kärkkäinen. “The ONE Simulator for DTN Protocol Evaluation,” In
Proceedings of ICSTT, 2009.
[12] A. Vahdat, and D. Becker, “Epidemic Routing for Partially Connected Ad Hoc Networks,”
Department of Computer Science, Duke University, Technical Report, April 2000.
[13] A. Lindgren, A. Doria, and O. Scheln, “Probabilistic Routing in Intermittently Connected
Networks,” Proceedings of ACM International Symposium on MobiHoc, vol. 7, no. 3, 2003, pp.
19–20.
[14] J. Burgess, B. Gallagher, D. Jensen, and B.N. Levine, “MaxProp: Routing for Vehicle-Based
Disruption-Tolerant Networks,” In Proceedings of IEEE INFOCOM, Barcelona, Spain, April
2006, pp. 1–11.
[15] A. Balasubramanian, B.N. Levine and A. Venkataramani, “DTN Routing as a Resource Allocation
Problem,” In Proceedings of ACM SIGCOMM, Kyoto, Japan, August 2007, pp. 373–384.
 Simulation Tools for Opportunistic Networks  101

[16] T. Spyropoulos, K. Psounis and C.S. Raghavendra, “Spray and Wait: an Efficient Routing Scheme
for Intermittently Connected Mobile Networks,” In Proceedings of ACM WDTN, Philadelphia,
PA, August 2005, pp. 252–259.
[17] T. Spyropoulos, K. Psounis and C. S. Raghavendra, “Spray and Focus: Efficient Mobility-Assisted
Routing for Heterogeneous and Correlated Mobility,” In Proceedings of IEEE Per. Com, White
Plains, NY, March 2007, pp. 79–85.
[18] M.S. Hossen and M.S. Rahim, “Impact of Mobile Nodes for Few Mobility Models on Delay-
Tolerant Network Routing Protocols,” International Conference on Networking Systems and
Security (NSysS), ACM IEEE, Bangladesh University of Engineering and Technology, Bangladesh,
January 2016.
[19] Project page of ONE simulator. Available: http://www.netlab.tkk.fi/tutkimus/dtn/theone, 2013.
[20] F. Bai, N. Sadagopan, and A. Helmy, “IMPORTANT: A Framework to Systematically Analyze the
Impact of Mobility on Performance of Routing Protocols for Ad hoc Networks,” In Proceedings
of IEEE INFOCOM 2003, vol. 2, March 2003, pp. 825–835.
[21] T. Issariyakul, An Introduction to Network Simulator NS2, New York, NY: Springer.
Chapter 6

Understanding Influencers of Adaptive

Social-Aware Opportunistic Forwarding
Soumaia A. Al Ayyat and Sherif G. Aly
American University in Cairo, Cairo, Egypt

CONTENTS

6.1 Introduction............................................................................................................. 104

6.2 Background and Related Work................................................................................ 105
6.2.1 Context-Adaptive Forwarding in Mobile Opportunistic Networks............... 105
6.2.2 EBubbleRap Algorithm................................................................................. 105
6.2.3 PeopleRank Algorithm ................................................................................. 106
6.2.4 SCAR Algorithm........................................................................................... 106
6.2.5 PI-SOFA Framework..................................................................................... 107
6.2.5.1 PIPeROp Algorithm....................................................................... 108
6.2.5.2 PISCAROp Algorithm.................................................................... 109
6.2.6 Space Syntax................................................................................................. 110
6.2.6.1 Space Syntax Metrics...................................................................... 110
6.2.6.2 Space Syntax-Based Forwarding Algorithms ................................... 111
6.3 Adaptive Ranking.................................................................................................... 113
6.3.1 Adaptive Ranking Framework...................................................................... 114
6.3.2 Adaptive Ranking Versions........................................................................... 118
6.4 Adaptation Proposed Implementations.................................................................... 119
6.4.1 Version 1: Adp .............................................................................................119
6.4.2 Version 2: AdpOp......................................................................................... 122
6.4.3 Version 3: AdpSyn ........................................................................................122
6.4.4 Version 4: AdpSpSynOp................................................................................ 123
6.5 Evaluation ..............................................................................................................123
6.5.1 Simulation Environment ..............................................................................124
6.5.2 Evaluation Metrics........................................................................................ 126
6.5.2.1 Effectiveness................................................................................... 126
6.5.2.2 Efficiency........................................................................................ 126
6.5.2.3 Power Awareness............................................................................ 127
6.5.2.4 Normalized Performance Indices .................................................... 127
6.6 Results..................................................................................................................... 128
6.6.1 Interest Awareness......................................................................................... 128
6.6.1.1 Effectiveness ................................................................................... 128
6.6.1.2 Efficiency........................................................................................ 130
6.6.2 Power Awareness.......................................................................................... 130
6.6.2.1 Power Consumption Awareness...................................................... 131
6.6.2.2 Power Utilization Fairness.............................................................. 132
6.6.2.3 Overhead of Exchanged Control Messages..................................... 132
6.6.3 Normalized Performance Indices................................................................... 134

103
104  Opportunistic Networks

6.6.3.1 Effectiveness Performance Index..................................................... 134

6.6.3.2 Efficiency Performance Index.......................................................... 135
6.6.3.3 Power Awareness Performance Index .............................................. 135
6.6.4 Eight-Metric Performance Comparison......................................................... 135
6.7 Discussion and Conclusion .....................................................................................136
Acknowledgment ............................................................................................................137
References .......................................................................................................................137

6.1 INTRODUCTION

Social-based opportunistic forwarding algorithms are essential for content dissemination in

environments with insufficient network infrastructure or in those that are prone to frequent
disruptions [1]. These forwarding algorithms leverage the knowledge of the users’ social
profiles in ranking the nodes for forwarding decision making [2]. However, many of these
algorithms are insensible to both the user’s interest in the forwarded content [3] and the lim-
ited power resources of the available mobile nodes [4]. Very few of these algorithms [5,6] are
aware of these two significant pieces of context information when ranking candidates, yet
they are oblivious to the importance of adapting their ranking algorithm to the dynamically
changing surrounding environment, as discussed in Section 6.2. For instance, as the message
is being transferred from one environment to another, some context information might be
inaccessible or may even become less important in the new environment. Due to this change
in the availability or the degree of importance of context information, over time, an adaptive
approach of ranking candidates for forwarding decision making is much needed.
Adaptive ranking is a framework that dynamically integrates a set of parameters to com-
pose a rank for each node. The forwarding decision making relies mainly on this computed
rank in order to adapt to the current context. The main proposed parameters of the rank are
weighted attributes which include opportunistic selection of forwarders among the surround-
ing candidates, the user interest and their mobile node power capability, a measure of how
active socially popular users are, and the popularity of the places these users frequent. With
such changing rank, the nodes are continuously re-evaluated for the sake of optimizing the
achievement in terms of delivery ratio, cost, delay, power consumption, utilization fairness,
and f-measure. This framework is defined in Section 6.3.
Space Syntax is a set of techniques that analyze spatial configuration based on theories
that link space and society relating spatial configuration to where and how people move and
adapt to space. Few contributions have introduced Space Syntax in the domain of opportu-
nistic forwarding [7–9]. In this work, we introduce the integration of Space Syntax into the
adaptive ranking framework. Section 6.4 illustrates four proposed implementations of this
framework that cater to the adaptive weighing of the attributes. The first version is the basic
adaptive algorithm, while the second version additionally introduces opportunistic selec-
tion of forwarders. The third and fourth versions of the adaptive algorithm integrate Space
Syntax metrics in the adaptive ranking functions of versions 1 and 2, respectively.
Section 6.5 evaluates the performance of the proposed adaptive ranking implementations
through simulations based on the SAROS simulator [10], which is introduced in previous
research by the authors of this chapter. This simulator includes various environment context-
aware simulations that are significant for simulating the forwarding process using the four
proposed implementations. The simulator also simulates the forwarding process using the
selected related work implementations; namely, the Epidemic, and the state-of-the-art social-
aware opportunistic forwarding algorithms which include some sort of adaptation, such as
the EBubbleRap, the PIPeROp and the PISCAROp algorithms [11]. The simulated context
information includes mobility traces from a university campus, power consumption, interest
 Social-Aware Opportunistic Forwarding  105

awareness, social profiles, and Space Syntax awareness metrics computation. The results
of these simulations are discussed in detail in Section 6.6. The results are quite promising,
as they demonstrate improvement in forwarding performance of the proposed algorithms
compared to the aforementioned leading-edge social-aware opportunistic forwarding algo-
rithms. For instance, the simulation results illustrate that the first proposed version outper-
forms the benchmark algorithm, Epidemic, in addition to the state-of-the-art EBubbleRap,
PIPeROp, and PISCAROp algorithms by a 40% increase in f-measure, a 96% reduction in
the ratio of contacted uninterested users, a 75.6% reduction in cost, and a 9% reduction
in power consumption per unit delivery. Comparing the performance of the four proposed
versions to one another, it is found that the first version achieves the highest ratio of contact-
ing interested users, while the fourth version contacted the least ratio of interested users.
However, this fourth version achieved the highest f-measure among all the algorithms. In
addition, the third version incurred the least cost and least power consumption to achieve
the highest delivery ratio among the four proposed versions. Finally, we conclude with some
recommendations of the most suitable forwarding algorithm implementation(s) for various
real-life application environments.

6.2 BACKGROUND AND RELATED WORK

There is significance in dynamically adjusting the weights of the attributes of the ranking
function due to the fact that the surrounding context that the nodes encounter along their
path in opportunistic networks is continuously varying. Surveying the literature, very few
contributions are found to be aware of the importance of dynamically adjusting the weights
of the attributes of the ranking function. In this section, we discuss the few approaches that
made partial steps towards providing dynamic adjustment of the weights of the ranking func-
tion’s attributes.

6.2.1 Context-Adaptive Forwarding in Mobile Opportunistic Networks

The research paper [12] has shown that for achieving a more efficient routing in opportunis-
tic networks, it is recommended to compute the weights of the ranking function parameters
based on the node’s behavior rather than to set predefined weights. The authors proposed
setting weights of an opportunistic routing utility function per group of nodes to improve the
performance of the network in terms of hit rate, latency, and delivery cost. They also showed
through simulations that dynamic adjustment of the weights of the parameters that are simi-
lar in value further improves the network performance. However, they propose algorithms
that remove the attribute in which two encountering nodes maintain the same value, and
thus, the algorithm redistributes the weight of this attribute among the rest of the attributes.
They also did not set power-aware attributes to dynamically adapt with the scarceness of
power resources in the mobiles nodes.

6.2.2 
EBubbleRap Algorithm
The community-aware Bubble Rap [13] routing algorithm cares for the fact that nodes’ rank-
ing differs as they move within their local community from their global ranking within the
whole community. Thus, this algorithm sets two ranking functions. These ranking functions
are the local ranking function that ranks the nodes within their local communities, and the
global ranking function that ranks the nodes within the global community.
The EBubbleRap algorithm [14] is the energy-aware version of the Bubble Rap routing
algorithm. This version introduces energy consumption optimization to the socially aware
routing [14]. The EBubbleRap algorithm introduces an energy-aware utility function into
106  Opportunistic Networks

the local and global ranking functions that are used in the BubbleRap algorithm. However,
being insensible to forwarder’s interest in the content, the algorithm contacts uninterested
nodes overloading them with content not in their interest domain and also consumes their
resources. We would like to mention that throughout the chapter, the term nodes’ interest
maps to the user’s interest.

6.2.3 
PeopleRank Algorithm
The PeopleRank algorithm is a power-oblivious, social-aware message forwarding algorithm
based on forwarding content by utilizing the nodes of the socially popular people in a par-
ticular place [15]. The PeopleRank algorithm is based on the assumption that socially popu-
lar nodes represent better candidates for messages delivery to destinations since these nodes
with higher probability will encounter destinations more quickly. Social ranking of nodes is
based on their social relationship. Such social relationship is recognized through a declared
friendship, or through shared common interests among the nodes.
The PeopleRank algorithm achieves a balance between social-based forwarding and
opportunistic forwarding through introducing a damping factor for deciding the ratio of
reliance on the social ranking of a node versus the reliance on the opportunistic encounter of
carriers. The damping factor (d) ranges from a value of 0, for total reliance on opportunistic
forwarding, to 1 for total reliance on social-based forwarding. Values in between represent
the relative weight set for each of the two approaches.
The researchers proposed a contact-aware version of the PeopleRank algorithm (CA-PeR)
that ranks nodes using both their social rank and their degree of social activeness. Node
activeness is measured by how frequent it comes across its social contacts. In effect, a node’s
social rank is either rewarded or penalized according to a count of the node’s contacts with
its social contacts. The CA-PeR rank as defined in [15] is computed as follows:

CA  PeR  j   wi , j
CA  PeR  i   1  d   d    | F j |
(6.1)
jF i

where d is the damping factor, CA − PeR(j) is the PeopleRank value of the friend j, F(i) is the
set of friends of node i who are available in vicinity, and wi, j is the contact-component and
computed by the following equation:

| encountersi , j |
wi , j  (6.2)
 | encountersi ,k |
kF  i 

The CA-PeR value consists of a pure opportunistic component (1 − d), a contact-aware

component w, and a social-aware ranking component. The social-aware component is based
on the concept that the node has a high social rank if, on average, its friends have high social
ranks. This component is computed by calculating the sum of the CA-PeR values of all the
friends j of the node i that is then averaged by the count of those friends ∣F(i)∣.

6.2.4 SCAR Algorithm
SCAR [6] is a sensor context-aware opportunistic routing protocol. The SCAR algorithm is
an adaptive power-aware version of the SocialCast algorithm [16]. SCAR is suitable for envi-
ronments where mobility patterns can be predicted to a certain extent, such as forwarding
 Social-Aware Opportunistic Forwarding  107

sensor data through existing routes among mobile sensor nodes to reach fixed or mobile des-
tination nodes. This protocol is aware of the power consumption and selects the route that
consumes less power. The SCAR protocol selects the best candidates for forwarding sensor
data based on forecasting the likelihood that they will meet destination nodes. The proto-
col is distributed and preserves information privacy since each node computes its attributes
locally. To improve performance, the SCAR algorithm relies on the anticipated values of the
utility attributes instead of on their current values. The forecast mechanism observes three
factors: history of co-location of the candidate node with the destination nodes, the change
of degree of connectivity of the candidate node with other sensor nodes, and the level of
battery of this node. The SCAR algorithm includes a monotonically decreasing function that
adapts to the predefined ranges of values of a certain attribute. The utility function weighs all
the attributes so that it maximizes the benefit gained from them all, and in order to balance
the trade-off between all of these attributes. The route selection process is based on selecting
those sensor nodes with the higher utility value from among their neighbor sensors.
The node’s utility rank as defined in [6] is shown in Equation 6.3.

Util  i   Rangecol  wcol  Pcol  i   Rangecdc  wcdc  Pcdc  i 

(6.3)
 Rangebat  wbat  Pbat  i 

where Pcol(i), Pcdc(i), and Pbat(i), respectively, are the predicted co-location of node i with
any of the destination nodes, the predicted change degree of connectivity of node i, and the
predicted remaining battery level of node i. The three predicted attributes are forecasted by
applying the Kalman Filter prediction techniques. The weights wcol, wcdc, and wbat, respec-
tively, represent the relative importance of the co-location attribute, the change of degree
of connectivity attribute, and the battery level attribute. The authors of the SCAR contribu-
tion mention that these weights’ values depend on the application scenario. Also, Rangecol,
Rangecdc, and Rangebat are the adaptive monotonically decreasing range functions for the
respective attributes.

6.2.5 PI-SOFA Framework
In another research [11], we presented our vision of a framework that enables integrating
interest and power awareness both in ranking the nodes and in the decision process within
social-aware opportunistic forwarding algorithms. This framework is applicable for social-
aware opportunistic forwarding in one-to-one hop mobile opportunistic networks in urban
areas where direct Wi-Fi or Bluetooth connection among mobile nodes enables the message
forward for a soft real-time delivery within a couple of hours.
Integration of interest awareness in social-aware forwarding algorithms is simply based
on rewarding (or penalizing) the social rank of the nodes based on their potential interest
in the forwarded message. The nodes that have potential interest in the forwarded content
are those whose interest profiles partial commonality with that of the forwarded message.
Both interest profiles can be represented in the form of an interest vector; then both vectors
are compared using the Jaccard set similarity Index [17] to coin a new measure, namely, the
Similarity Interest (SInt).
To be fair in utilizing the power of the nodes in the place, the social-aware forwarding
algorithm has to be aware of two pieces of information, namely, the current power resources
of the nodes in the place, and the expected power consumption upon their participation in
the message forward process. With such knowledge, the algorithm favors the nodes with
higher power capabilities that can sustain the delivery until completion and avoid exhausting
the nodes with poor power capabilities.
108  Opportunistic Networks

Within this framework, several implementations of PI-SOFA algorithms that integrate

interest and power awareness into the state-of-the-art social-ware opportunistic forwarding
algorithms are proposed such as the PIPeROp algorithm, the PISCAROp algorithm, and the
PISCast algorithm. In this section, we will detail two of these proposed algorithms.

PIPeROp Algorithm
6.2.5.1 
The PIPeROp algorithm [5] is an interest and power aware social opportunistic forwarding
algorithm within the PI-SOFA framework. The PIPeROp algorithm adds two new dimensions
to the PeopleRank algorithm [15] by integrating power awareness and interest awareness
into the social-based forwarding process. In order to consider a mobile node for forwarding
a message, the PIPeROp algorithm elicits the candidate node’s available battery level and
the similarity in interest between the node and the forwarded content. The algorithm uses
this information as a means of indicating the node’s willingness to forward. In addition, it
emphasizes power awareness by rewarding the node that is more power abundant; other-
wise, it penalizes the node’s rank. Accordingly, a candidate node is highly ranked when the
following conditions are met: its battery exceeds a pre-defined threshold, its owner is socially
connected to popular friends, and both the node owner and their first-order social contacts
have interest in the message being sent. Thus, the higher the rank of a mobile node and its
social contacts, the higher the chance a node will become a candidate for forwarding the mes-
sage. However, both algorithms, PeopleRank and PIPeROp, rely on a damping factor that is
assigned a predefined value that is static per forwarding application. Such static value does
not adapt with the changes in the surrounding environment or the changes in the availability
of context information. The node’s power-aware rank, PIPeR, is formalized by this equation:

PIPeR  i   Re wardedPower  i   IPeR  i 

 PeR  j   wi , j 
  1  d   d  FrInter  i, msg  
 
jF  i 
| F j |


(6.4)

 

where RewardedPower is the power of the node being rewarded (or penalized) based on
comparing its value against a predefined battery threshold Thrbat as detailed in Equation 6.5.
Also, d is the damping factor, and FrInter(i, msg) is the cumulative interest of node i and of
its friends in the message content (it is calculated as per Equation 6.7), F(i) is the set of friends
of node i who are available in vicinity, PeR(j) is the PeopleRank value of the friend j, wi,j is
the weight of introducing the friend’s PeopleRank value based on the frequency of contact
between the two nodes.

RewardedPower  i  
(Bat  i   reward) if Bat  i   Thrbat , (6.5)

 Bat  i   reward  otherwise

Note that the IPeR value is the integration of interest-awareness with the PeopleRank
algorithm and is computed as per Equation 6.6:

IPeR  i   1  d  
CA-PeR  j   wi , j
 
(6.6)
d  FrInter  i, msg  
| F j |
jF i
 Social-Aware Opportunistic Forwarding  109

FrInter  i, msg  
PSInt  i, msg    PSInt  j, msg 
jF  i 

| F  i  | 1
(6.7)

The PIPeROp algorithm [11] adds an additional opportunistic component into the can-
didate selection process. This component favors forwarding the message to any interested
forwarder whose battery level is above the fixed battery threshold Thrbat. These favored
forwarders need not be socially popular users but should be power-capable users with high
interest in the forwarded content. Thus, the next message forwarder will be any candidate
whose similarity interest with the message’s interest vector exceeds a preset interested for-
warder threshold and whose node power is above the given battery threshold. In addition,
the algorithm selects candidates with a power-aware rank that is not less than that of the
current message holder.
However, both the PeopleRank and PIPeROp algorithms rely on a damping factor that is
assigned a predefined value that is static per forwarding application.

6.2.5.2 PISCAROp Algorithm
The PI-SOFA framework integrates interest awareness and power awareness into the SCAR
algorithm to propose a new power-and-interest-aware version, the PISCAROp algorithm
[11]. The PISCAROp algorithm rewards or penalizes the adaptive SCAR utility value of a
node based on the collective interest of both the node’s owner and their friends in the for-
warded message. It also penalizes or rewards the utility function based on the remaining
power of the node compared to a predefined battery threshold. Accordingly, those partially
interested nodes that are power capable become favored forwarder candidates.
This version minimizes contact with uninterested ones and favors potentially interested
nodes that are socially connected to other potentially interested friends. In addition, through
this version, the partially interested nodes that are power capable become favored forwarder
candidates.
The new utility function is computed as follows:


lutil  i  
P

 PSInt  i, msg   RewardedPower  i   Util  i  ifPSInt  0 & RewardedPower  0,
(6.8)
 PSInt  i, msg   RewardedPower  i   Util  i  otheerwise

where PSInt(i, Ad) is the rewarded/penalized similarity interest of the node with the forwarded
content interest vector as detailed in Equation 6.9, RewardedPower(i) is the rewarded/penal-
ized battery level of the node that is computed as per Equation 6.5, and Util(i) is computed
as per Equation 6.3 of the SCAR algorithm.

(SInt  j, msg   reward if SInt  ThrInt ,

PSInt  j, msg    (6.9)
(SInt  j, msg   reward otherwise

All the above utility function computations are based on the preset interest and battery
thresholds thrInt and thrbat, respectively, that do not adapt with the changes in the surround-
ing context.
110  Opportunistic Networks

6.2.6 Space Syntax
Space Syntax was initially proposed in the field of architecture by Hillier and Hanxon in
1984 to model natural mobility patterns by analyzing spatial configurations. It accurately
predicts natural movement patterns in an area based on how its segments are connected [8,
18, 19]. Space Syntax is comprised of a set of techniques for analyzing spatial configuration,
and it is derived from a set of theories linking space and society. Using Space Syntax, one can
relate spatial configuration to where people are, how they move, and how they adapt to space
[18]. Space Syntax opportunistic forwarding exploits the infrastructure of a particular place
and the popularity of specific attraction points in that place in order to predict user mobility
patterns, flows, and areas of possible congestion [8, 9].

6.2.6.1 Space Syntax Metrics

In the domain of Space Syntax there are standard metrics to quantify the effect of road maps
and architectural configuration on natural movement. Such metrics predict natural move-
ment patterns in an area based on the way of segments connectivity. Space Syntax metrics
are pre-calculated once using static information about the environment without a need for
recalculation unless the environment changes [8]. There is a set of basic Space Syntax metrics
that are commonly used to set the popularity of any street or spot on a given map. We con-
centrate here on five basic metrics that are detailed as follows:

Integration value [20]: The segment’s integration value is the average distance between the
segment and all other segments in the map. This value represents the segment’s close-
ness to other segments.

IntegrationValue  x  
(6.10)
Average   y in Segments Distance  x, y  

Connectivity value [20]: The segment’s connectivity value is the number of segments inter-
secting with this segment. It represents the segment’s degree of connectivity.

ConnectivityValue  x    y in Segments Intersecting with x Count  y  (6.11)

Location index [8]: The location index is the deviation of the segment’s integration value
from the maximum integration value divided by the range of integration values.

Location Index  x    y in Segments

max  IntegrationValue  y    IntegrationValue  x  (6.12)

Range  IntegrationValue  y  

Attraction value [8]: It is an index that indicates the level of attraction of the spot with
respect to the rest of the spots in the area.

AttractionValue  x   w.r.t. all the Segments

(6.13)
Index  x 
 Social-Aware Opportunistic Forwarding  111

Popularity index [8]: The popularity index of a segment is calculated as the normalized
attraction value of the segment added to the sum of the normalized length indexes mul-
tiplied by the normalized attraction values of all the segments that intersect with this
segment. The normalized length of an intersecting segment is the length of the intersect-
ing segment divided by the segment whose popularity index is being calculated.

Popularity Index  x   Normalize  AttractionValue  x  

  y in Segments Intersecting with x (6.14)
 Length  y  
  Normalize  AttractionValue  y   
 Length  x  
 

6.2.6.2 Space Syntax-Based Forwarding Algorithms

We located a number of studies that looked at Space Syntax specifically in terms of how it
affects or controls the social behavior of humans in a particular place and which inferred
human mobility and social interaction from these behavior patterns [20–23].
Some studies built on Space Syntax in order to improve the opportunistic forwarding pro-
cess, including the following:

Space Syntax forwarding by popularity index [8]: Building on the Space Syntax, these
authors introduce a new Space Syntax metric, namely, the popularity index of any
attraction point on the map. This new metric is computed as a function of the segment’s
integration value, its location index, and its attraction value. Based on the computed
segments’ popularity indexes, the popularity index of any attraction point on the map
is then calculated. However, the limitation of their proposal is that it relies on the condi-
tion of first finding and establishing a route to the destination node before initiating the
forwarding process. Satisfying this condition could be difficult in disrupted networks
and thus limits the area of application of their proposal. The authors demonstrate the
effect of the newly defined metric by proposing the following two Space Syntax-based
forwarding algorithms:
• The SS-Cutoff algorithm prohibits message forwarding to nodes that are geographi-
cally located in areas whose popularity falls between a certain configurable thresh-
old. Also, with a given probability, it forwards messages to nodes in areas whose
popularity is above or equal to the threshold.
• The SS-Sliding algorithm forwards messages to nodes according to a variable prob-
ability that is inversely proportional to the popularity of the area in which these
nodes reside. Thus, forwarding becomes less aggressive in areas that are more popu-
lar so as to decrease cost.

Select & Spray [19]: This algorithm is a Space Syntax-based opportunistic forwarding
algorithm that is applicable on a citywide level. The authors consider Space Syntax
information to guide opportunistic forwarding based on the likelihood of a location to
attract mobile nodes. The approach also depends on a server/client setup, not an oppor-
tunistic network setup. The server consists of a “data collector”, a “recommendation
engine”, and a “select engine” that collect and memorize registered clients’ mobility
data in addition to a priori knowledge of the city map and the position of attraction
points. The algorithm then computes a maximum probability utility function which
112  Opportunistic Networks

ranks all nodes in the network based on their probability of reaching a destination,
and then it selects the best subset of connected nodes that will be used to disseminate a
message with the least potential cost and delay. Based on a citywide range and a server/
client configuration, this application is rather costly in terms of computations, com-
munications, and resources.
The data collector fetches the data of registered clients and the surrounding envi-
ronment, then inserts or updates them in the databases. The recommendation engine
consists of recommendation algorithms to recommend messages to interested clients.
It initiates communications by sending the message and the set of destinations to the
select engine. The select engine relies on a priori knowledge of the city map, the posi-
tion of attraction points, and/or the history of node mobility patterns. The select engine
has to select the best subset of connected nodes that will be used to disseminate a mes-
sage with the least potential cost and delay. The client mobile nodes implement a spray
engine. This engine consists of an opportunistic dissemination algorithm with forward-
ing decisions based on social, historical, or mobility data. The select engine detects the
probability of divergence between two nodes at time t + 1 to eliminate outliers. There is
a maximum probability utility function which ranks all nodes in the network based on
their probability to reach a destination.
Exploiting Space Syntax for Mobile Opportunistic Networks: In a study that exploits
Space Syntax for Mobile Opportunistic Networks, Mitbaa et al. [9] survey the field
of opportunistic forwarding in small-scale urban environments and categorize them
according to the assumptions on which the algorithms base the forwarding decisions.
The authors illustrate that these algorithms – as they rely on more complex assump-
tions – achieve more efficient performance, “rendering solutions built on them unusable
outside their intended environment categories” [9]. This study proposes three sample
algorithms, each of which represents one of their defined categories. The proposed algo-
rithms assume that the forwarded messages may not be fragmented, and that they are
designed to ensure there is enough time for nodes to exchange messages. The forward-
ing decisions are based on static information, such as city map and attraction points,
or, alternatively, on local information regarding the node itself. The authors propose the
following three sample algorithms:
• Space Syntax Opportunistic Forwarding (SOF): This forwarding approach is based
on the nodes’ awareness of a map layout.
• Location-Aware Space Syntax Opportunistic Forwarding (LA-SOF): This forward-
ing approach is based on the nodes’ awareness of a map layout, the nodes’ position,
and the position of the destination Internet access point locations.
• Assisted Space Syntax Opportunistic Forwarding (ASOF): This forwarding appro­
ach is based on the nodes’ awareness of a map layout, the nodes’ position, the
position of the destination Internet access point locations, and the nodes’ mobility
patterns.

Space Syntax and pervasive systems: In a study of how innovative the application of Space
Syntax in the domain of pervasive systems can be, the author applied Space Syntax
metrics in different phases of the development of pervasive systems [20]. The author
proposes three case studies that demonstrate the role of Space Syntax as a simple appli-
cation module, as an explanatory tool, and as a modeling tool.
The Weighted PageRank approach [22] is another study that proves that the underlying
space–space topology of an urban environment is far from random. Rather, it exhibits
small-world and scale-free structures and proves that the weighted PageRank approach
can accurately predict human movement in an urban environment.
 Social-Aware Opportunistic Forwarding  113

All the aforementioned research contributions exhibit the following limitations:

First, the few contributions that combine Space Syntax metrics with other opportunistic
forwarding techniques mainly consider expansive areas at a citywide level. They do
not consider small scale level application such as a campus-wide or convention center
application.
Second, some of these contributions consider social graph metrics (such as closeness
centrality and betweenness centrality) in addition to the contact graph, but no proposed
algorithm so far considers the forwarder nodes’ interest in the forwarded content.
Third, the ASOF approach and its peer algorithms consider destination locations such
as streets or static access points [9, 19] – except for the Popularity Index approach
which considers mobile destination nodes [8]. Thus, their proposed algorithms cannot
be applied to mobile destination nodes.
Fourth, all the proposed Space Syntax-based contributions to date do not take into
consideration power consumption, and some complex algorithms consume a great deal
of energy. The preservation of energy is crucial in limited-resources mobile opportunistic
networks.
Finally, the proposed algorithms by Mitbaa et al. [9, 19] rely on the concept that the mobile
node continuously recomputes the weights of its own contact graph to re-compute its
utility function. Practically speaking, these computations consume the node’s resources
especially when this approach is applied on a citywide scope. It is worth noting that
the node needs to compute this graph for each forwarded message since each message
has its own set of destination locations or destination access points. Given that most
of the messages will be forwarded through the betweenness-central mobile nodes, this
computational consumption leads to unfairness by exhausting the node’s resources.

6.3 ADAPTIVE RANKING

In many cases, there is a frequent dynamic change in the available set of context information
needed for ranking nodes in order for the algorithms to properly select the best candidates
for forwarding. Thus, dynamic weights of the attributes are needed in order to compensate
for that change in context. For this purpose, this research proposes an adaptive ranking algo-
rithm that integrates five context-aware sensors, namely: interest awareness, power aware-
ness, and place awareness with social awareness and user activeness. The algorithm takes
into consideration various factors that affect the rank of the coexisting nodes. These factors
include the following:

• The relative interest of the user in the forwarded content;

• The power capabilities of the mobile node the user holds;
•
The degree of social popularity of the user;
•
The degree of activeness of the user in terms of physically or logically contacting nodes
of other users;
• The popularity of the places the user frequents.

This approach dynamically changes the weights of the aforementioned factors to suit the sur-
rounding environment. Based on these changing weights, the factors that control the node’s
rank are dynamically weighed to adapt the changing environments the node moves across.
With such changing rank, the nodes are continuously re-evaluated for the sake of optimizing
the achievement of the forwarding algorithm in terms of the following metrics:
114  Opportunistic Networks

• Increasing the delivery ratio;

• Increasing the contact with the interested users;
• Minimizing the contact with the uninterested users;
• Minimizing the cost paid while the targeted delivery ratio is achieved;
• Minimizing the amount of consumed power while the targeted delivery ratio is reached;
• Maintaining fairness in utilizing the resources of the available mobile nodes in the
place.

Accordingly, each node will execute this ranking function utilizing its own values of the
aforementioned factors. Its rank changes as the weights of these factors change; thus, with
each new encounter, the node recomputes its own rank and exchanges its value with the
encountered node for decision making on content forwarding.

6.3.1 Adaptive Ranking Framework

In this framework, the forwarding decision relies mainly on the dynamic rank in order to
adapt to the current context. The main proposed parameters of the rank are dynamically
weighted attributes which include: opportunistic selection; the node’s owner interest and
the node’s power capability; the measure of how active socially popular users are; and the
popularity of the place these users frequent. This takes into consideration that as nodes move
from one place to another, or as time passes, both the value and the relative importance of
their attributes may change. Then, a fixed weight of each attribute within the node’s rank
might not reflect the current context, leading to an incorrect node ranking approach. Hence,
continuously re-evaluating the attributes’ weights helps in reflecting the correct current rank
of the node. Accordingly, this ranking framework adapts to the current context by changing
the relative weights of the attributes that decide the node’s rank. The change of weights relies
on the values of the aforementioned attributes and whether these current values fall into
critical ranges that set such attributes ineffective in comparison to the remaining attributes at
the current context. Other changing contexts would take place where any of these attributes’
values are unavailable at any certain time due to any reason, or because the user or their
mobile node refuses to reveal its value.
Based on this logic, the node ranking function ranks candidates based on dynamically
weighted attributes which are enclosed in the following components:

. The opportunistic selection (OppComponent);

1
2. The user’s interest (IntComponent);
3. The measure of how active the socially popular users are (SocActiveComponent);
4. The power capability of the mobile node (PowerComponent);
5. The popularity of the places frequently visited by these users (PlaceAwareComponent).

This adaptive ranking function can be formalized as follows:

AdpRank  i   Oppcomponent  IntComponent

 SocActiveComponent  PowerComponent (6.15)
 PlaceAwareComponent

Let us have a closer look at the various components.

 Social-Aware Opportunistic Forwarding  115

1. The opportunistic forwarder node selection component, OppComponent, gives space

for a pure opportunistic selection of candidates, OpportunisticSelection, without being
restricted by any certain criteria. This component is allowed to take place with a wOpp
portion of the ranking function components. This component can be formulated as
follows:

OppComponent  wOpp  OpportunisticSelection (6.16)

2. The user’s interest component, IntComponent, considers the relative interest of the user
in the forwarded content as an important factor in ranking nodes. This component
consists of the similarity in interest between the user’s interest vector and the forwarded
content domain of interest SInt(i, msg). This similarity interest metric can be com-
puted using Jaccard similarity between the two interest feature vectors [17]. It is more
comprehensive to include the similarity interest of the user’s friends CumInterest(i) as
well. Another interesting component that might be considered in the user’s interest is
the power capabilities of the mobile node they are carrying, Power(i), since the power
limitation of some devices might hinder the user’s interest in receiving any content even
if it is within the user’s range of interest. This component can be formulated as follows:

IntComponent  wIntPower  CumInterest  i   Power  i  (6.17)

3. The social active component, SocActiveComponent, takes into consideration the social
popularity of this user among their friends. It also considers the level at which the
mobile user is active in contacting and interacting with other nodes. This component is
aware of the active members and the sociable members since this factor can be a good
indicator that such members will, with high probability, get in contact giving a higher
chance to meet with the destination nodes [24] and to forward the content to anyone of
the interested friends. This component consists of the SocialRank(i) of the node and the
measure of activeness of the node Activeness(i). There are several ways to measure these
two parameters. Accordingly, this component can have various implementations. The
ranking function relies on this component with the relative weight wSocActive. Generally
speaking, this component is formulated as follows:

SocActiveComponent  wSocActive  SocialRank  i   Activeness  i  (6.18)

4. The power capability of the mobile node component, PowerComponent, considers the
node’s power capability, Power(i), as an affecting factor in selecting the next forwarder.
In fact, the user whose mobile node’s power capability is not high may refrain from
participating in the forwarding process, and also the node that has low power capa-
bility may not survive until the completion of the delivery process. It is more effective
to highly rank the power capable node among all the available nodes in the place.
However, to maintain fairness in utilizing the nodes in the place, there is a need to set
a threshold below which nodes are forced to drop the forwarded message copy and
abandon the participation in the forwarding process. The power capability of a node
can be represented by various measures such as the remaining battery charge or by con-
sidering the depletion rate of the battery in addition to its capacity. Generally speaking,
this component is formulated as follows:
116  Opportunistic Networks

PowerComponent  wPowerAware  Power  i  (6.19)

5. The popularity of the place is an important component to consider in ranking nodes.

This component, namely, PlaceAwareComponent, relies on the importance of the
Space Syntax metrics on decision making. There are various Space Syntax metrics
that can be utilized in computing the place popularity attribute PlacePopularity(i).
Some of the popular metrics are the integration value, location index, and attraction
value, among others [8]. These metrics consider closeness centrality and between-
ness centrality among other social metrics in ranking places in terms of popularity
[18]. In a previous work, we have proposed variations of the Space Syntax metrics
[7]. The ranking function relies on the place popularity component with the ratio
computed by the place popularity weight wPlacePopularity. This component is formulated
as follows:

PlaceAwareComponent  wPlace Popularity  Place Popularity  i  (6.20)

where PlacePopularity(i) computes the place popularity of node i using Space Syntax met-
rics such as location popularity of the places frequently visited by node i, or the location pop-
ularity of the places node i stays the longest time duration among other Space Syntax metrics.
In greater detail, the adaptive ranking function is computed as:

AdpRank  i   wOpp  OpportuniticSelection

 wIntPower  CumInterest  i   Power  i 
(6.21)
 wSocActive  SocialRank  i  Activeness  i 
 wPowerAware  Power  i   wPlacePopularity  PlacePopularity  i 

where wOpp = 1 − (wIntPower + wSocActive + wPowerAware + wPlacePopularity)

wIntPower   SInt  i, msg  , bat  i  

wSocActive  f  SInt  i, msg  

wPowerAware  f  bat  i  , depletionrate  i  , capacity  i  


wPlacePopularity  f Pop  Location  i, now   , Pop  FreqLocation  i   
There are five weights that dynamically change in value to adapt to the current context.
These weights are:

1. wOpp is the weight of the opportunistic selection component. Its value is actually the
remaining portion after computing all the weights of the other components.
2. wIntPower is a monotonically decreasing function of the similarity interest (SInt) of the
node and its power capability Power(i). The Power(i) parameter is the remaining power
of node i. This power component can be represented as the rewarded predicted power
of the node which is computed as RewardedPower(i) as per Equation 6.5. The battery
 Social-Aware Opportunistic Forwarding  117

depletion rate awareness can be also introduced in the ranking function where Power(i)
is computed as per PredPower(i), detailed in Equation 6.22.

Pr edPower  i  
 Bat  i 
 Bat  i   reward  if  TTL  msg   Thrdep, (6.22)
 DepRate  i 
(Bat i  reward) otherwise
 

where DepRate(i) is the depletion rate of node i, TTL(msg) is the time-to-live of

the forwarded message, and Thrdep is the threshold of the battery depletion rate below
which Bat(i) is penalized.
In some applications the users’ interest in the forwarded content may vary or decrease
as time passes. This can be represented by the similarity interest between the interest
vector of the user, IntFV (user), and the interest vector of the message content, IntFV
(msg), that may not be constant as it may change over time. Accordingly, the SInt(user,
msg) becomes a function of time. This is depicted in the following formula:

SInt  user, msg   f  time, IntFV  user  , IntFV  msg   (6.23)

3. wSocActive is the weight of the socio-activeness of the node. It is a function of the SInt of
the node. This is attributed to the fact that the ranking function pays more attention to
nodes that have interest in the forwarded content. That is why the node’s interest in the
forwarded content emphasizes its social and activeness components.
4. wPowerAware is the weight that sets the relative effect of the node’s power capability in the
node ranking process. This weight decreases as the node’s power decreases to indicate
the decreasing value of this node when selecting it to participate in the forwarding
process.
5. wPlacePopularity is the weight of the place popularity PlacePopularity(i) of node i. The
weight wPlacePopularity is a function of both the popularity of the current location of node i,
Pop(Location(i, now)), and the popularity of the most frequent location in which node i
stays along its path Pop(FreqLocation(i)). By the way, it may be replaced by other space
popularity metrics.

Let us clarify with a simple example. Node i will calculate its adaptive rank AdpRank(i)
based on these given values: the current remaining portion of the battery of node i is 0.8 (i.e.
80%), its depletion rate is 0.01 per second, and the capacity of this battery is 6000 mAh.
Then the weight of power awareness, wPowerAware, is a function f(0.8, 0.01, 6000).
Let the user of this mobile node be a sociable person with a social rank 0.9 as per one of
the state-of-the-art social ranking functions, and his level of activeness in communicating
with others is 0.7. His interest in the content of the forwarded message is 0.75 as reflected
by the similarity between the user’s interest vector and the interest vector of the forwarded
message, and the cumulative interest of this user and his friends in this message is 0.6 as per
the equation 1.38 that is detailed in the next section. Let the weight of the socio-activeness of
the user, wSocActive, be a function f(0.75, 0.7). Thus, the socio-active component will be f(0.75,
0.7) * 0.75 * 0.7.
Let the threshold above which the rewarded power of the node is rewarded be 0.5 and
let the reward value be 0.5. Then, the RewardedPower(i) is 0.8 + 0.5. From these values, the
118  Opportunistic Networks

weight of the interest-and-power awareness component, wIntPower, is calculated as a function

f(0.7, 1.3), while the interest-power component will be f(0.7, 1.3) * 0.6 * 0.8.
Now, let’s calculate the place popularity component. If the user frequently visits a place of
a popularity Pop(FreqLocation(i)) equal to 0.86, while he is currently located in a place of
popularity Pop(Location(i, now)) equal to 0.5, then, the weight of the place popularity of the
node, wPlacePopularity, is f(0.5, 0.86). The place popularity of the node is calculated as per one of
the Space Syntax metrics detailed in Section 6.2.6.1.
From all of these calculations, the weight of the opportunistic selection is wOpp = 1 −
(wIntPower + wSocActive + wPowerAware + wPlacePopularity). The opportunistic component itself would
be a random value of 1 (to select the candidate) or zero (to ignore this candidate). The
summation of all the aforementioned components constitute the adaptive rank of the node
as per Equation 6.21.

6.3.2 Adaptive Ranking Versions

The algorithms presented here adapt the ranking function according to the current context.
The value of the ranking function relies on several factors that include an interest and power
awareness factor, a social and activeness factor, a Space Syntax factor, and an opportunistic
forwarder selection factor. The implemented versions of this adaptive algorithm are:

1. The first version is the basic version of the algorithm which relies on the interest
and power awareness factor, the social and activeness factor, and the opportunistic
forwarder selection factor. This is summarized as:

AdpRank  i   f  IntPower  i  , SocActive  i  , Opp  i   (6.24)

2. The second version implements the adaptive algorithm and inserts an extra explicit
interest and power-based opportunistic selection criterion. Thus, the message carrier
node i selects node j as a forwarder if the following conditions are satisfied:

AdpRank  j   AdpRank  i  AND SInt  j, msg   Thrint AND bat  j   Thrbat (6.25)

3. The third version relies on the interest and power awareness factor, the social and
activeness factor, the Space Syntax factor, and the opportunistic forwarder selection
factor. It applies the same factors as the first version but with the Space Syntax factor
added to its ranking function. This is summarized as:

AdpSpSynRank  i   f (IntPower  i  , SocActive  i  , SpaceSyntax  i  , Opp  i ) (6.26)

4. The fourth version relies on the interest and power awareness factor, the social and
activeness factor, the Space Syntax factor, and the opportunistic forwarder selection
factor, then inserts an extra explicit interest- and power-based opportunistic selection
criterion. It applies the same factors as the second version but with the Space Syntax
factor added to its ranking function. Thus, a selected forwarder node j must satisfy the
following condition:
 Social-Aware Opportunistic Forwarding  119

AdpSpSyn Rank  j   AdpSpSyn Rank  i 

AND SInt  j, msg   Thrint (6.27)
AND bat  j   Thrbat

6.4 ADAPTATION PROPOSED IMPLEMENTATIONS

The surveyed works in the field of adaptive ranking and context-awareness seek social
awareness, power awareness, and space-popularity awareness, but they lack the ability to
adapt the weights of the components that compose these ranking functions. Through the
adaptive ranking approach, we present a ranking function that constitutes all the features
of social awareness, power awareness, and space-popularity awareness and also provides a
set of adaptive weights for all the components in order to be able to adapt continuously to
the current context. This section details the implementation of the four proposed versions
defined in Section 6.3.2.

6.4.1 Version 1: Adp
This is the basic version of the adaptive ranking function. This version basically considers the
following parameters in ranking nodes: the pure opportunistic component, the interest-and-
power-aware component, the social activeness component, and the node’s power component.
This ranking function is formulated as follows:

AdpRank  i   wOpp  Opportunistic

 wIntPower  CumInterest  i   Power  i 
(6.28)
 wSocActive  SocialRank  i   Activeness  i 
 wPowerAware  Power  i 

where the weights are computed as follows:

wOpp  1   wIntPower  wSocActive  wPowerAware 

wIntPower  SInt  i, msg   bat  i 

wSocActive  exp ectedusage  i 

wPowerAware  MonotonicallyDecreasing  Bat  i  

Note that the opportunistic component, Opportunistic, can be inclined to those users that
may be interested in the forwarded content and as well hold mobile devices with sustainable
power charge. It is then computed as follows:

Opportunistic  SInt  i, msg   Re mainingPower  i  (6.29)

where SInt(i, msg) is the similarity in interest between the node i and the forwarded message
msg, and RemainingPower(i) is the remaining battery level of the node i whose function is
illustrated in Equation 6.30;
120  Opportunistic Networks

Re mainingPower  i   f  battery  i  , exp ectedusage  i   (6.30)

where expectedusage(i) is the expected amount of power that will be consumed due to the
usage profile of the node i which indicates the average time consumed per activity. In addi-
tion, the power consumed due to network communications such as Wi-Fi scanning, forward-
ing, or receiving control messages is factored into the calculation of the expected usage.
While we are proposing the aforementioned method, we acknowledge that there may be sev-
eral methods of computing this component. For the simulation, whose results are illustrated
in this chapter, we implement the following version of the expected usage function:

ExpectedUsage  i   KalmanFilter( bat  i  ,

 
Avg Count Contacts  i   , TTL  msg  , (6.31)

Consumption WiF i  i  , FWD  i  , Rcp  i  , idle  i  )

This function predicts the expected amount of battery usage for a specified user – using
the Kalman filter predictor [25] – based on three factors. These factors are the node’s current
battery level bat(i), the expected power consumption due to future contacts as an extrapola-
tion from the average count of contacts recorded so far Avg(Count(Contacts(i))), and the
expected power consumption due to Wi-Fi connection WiFi(i) (or Bluetooth connection) and
actions of message forwarding or receiving (FWD(i) and Rcp(i)) from now until the expira-
tion time of the specified message time-to-live TTL(msg).
The social active component of a node, SocActive, is computed as follows:

SocActive  i   wSocActive  Activeness  i   SocialRank  i  (6.32)

where Activeness(i) is the measure of activeness of the node i in interacting with other nodes,
for which the formula is detailed in Equation 6.33; and SocialRank(i) is the social rank of
node i, which is detailed in Equation 6.36.
The activeness part of the SocActive component is computed as follows:


Activeness  i   1 / 3  normalizedContact  i, F  i    (6.33)
  1 / 3  CDC  i     1 / 3  Col  i, Dist  

where normalizedContact is the normalized count of contacts between node i and its set of
friends, F(i). CDC(i) is the change in degree of connectivity of node i, and Col(i, Dest) is the
normalized count of co-location of node i with any one of the destination nodes.
In this context, the change in degree of connectivity, CDC, is computed as the intersection
of the count of nodes co-located with node i at time t and those co-located with node i at time
t − 1, divided by the union of the count of those co-located with node i at time t and those
co-located with node i at time t − 1. As per its definition in the SocialCast research [16], CDC
is formulated as follows:

Count  col  m, i, t    Count   m, i, t  1 

CDC  i   (6.34)
Count  col  m, i, t    Count  col  m, i, t  1 
 Social-Aware Opportunistic Forwarding  121

where m is any one of the nodes co-located with node i at a certain time instance such that
the co-location of a node with any one of the destination node is:

Coli  i, DestinationSet  msg   

(6.35)

Normalize Dest  DestinationSet  msg  t Count Col  i, Dest   
The social rank of node i is computed as the normalized social rank of its friends as follows:

SocialRank  v 
SocialRank  i     | F v  |
(6.36)
vF i
Any of the social metrics may be applied in computing the social rank of a node such as
centrality or betweenness. In the first version, we combine the interest-aware and the power-
aware components to come up with the interest-and-power-aware component which is com-
puted as follows:

IntPowerComponent  i   Re wardedPower  i  
(6.37)
normalCumInt  i, msg 

where RewardedPower is computed as per Equation 6.5, while normalCumInt(i) is the cumu-
lative interest of the node which is computed as follows:

PInt  i   v  F  i  PInt  v 
normalCumInt  i   (6.38)
| F  i  | 1

where PInt(i) is the rewarded or penalized SInt(i) when its value is compared to the pre-
defined interest threshold.
The pseudocode of the algorithm is detailed in Algorithm 6.1

ALGORITHM 6.1  ADAPTIVE RANKING ALGORITHM

Require: SInt(source,   msg) = 0

1: ∀ time t every n seconds, ∀ nodes physically in contact

2: while node i is physically in contact with node j do
3: SocialRank(i) = Avg(∀υ ∈ F(i)  SocialRank(υ))
4: if j ∈ F(i) then
5: Update(SocialRank(i),   SocialRank(j))
6: end if
7: CalculateAdpRank(i)
8: if AdpRank(j) ≥ AdpRank(i)  AND TTL(msg)  > 0 then
9: Forward(msg, j)
10: end if
11: end while
122  Opportunistic Networks

Procedure CalculateAdpRank(i)
12: normalizedContact(i, F(i)) = Avg(∀υ ∈ F(i) | Contact(i, υ)| )
| Neighbors  i, now  |  | Neighbors  i, now  1 |
13: CDC  i  
| Neighbors  i, now  |  | Neighbors  i, now  1 |

14: Col(i, Dest) = Normalize(∀d ∈ Dest | Contact(i, d)| )

 
15: Activeness  i   1 / 3  normalizedContact  i, F  i     1 / 3  CDC  i     1 / 3 

Col  i, Dest  
16: RemainingPower(i) = f(bat(i),  expectedusage(i))

PInt  i, msg     F  i  PInt  ,msg 

17: normalCumInt  i, msg  
| F  i  | 1

18: SocActive(i) = Activeness(i) ∗ SocialRank(i)

19: wSocActive = expectedusage(i)
20: RewardedPower(i) = bat(i) ± reward
21: IntPowercomponent(i) = Re wardedPower(i) ∗ normalCumInt(i,  msg)
22: wIntPower = SInt(i,  msg) ∗ bat(i)
23: wOpp = 1 − (wIntPower + wSocActive)
24: Opportunistic(i) = SInt(i,  msg) ∗ RemainingPower(i)
25: AdpRank  i    wIntPower  IntPowercomponent  i     wSocActive  SocActive  i  

 wOpp  Opportunistic  i  

6.4.2 Version 2: AdpOp
This second version of the adaptive ranking algorithm, AdpOp, includes the option of select-
ing the next forwarder node if both its SInt(i, msg) and its remaining power bat(i) exceed
their respective predefined thresholds. This criterion of selection is added to the selection
criteria defined in the Adp algorithm detailed in Section 6.4.1.
Accordingly, a node is selected as the next forwarder if it satisfies the following conditions:

AdpRank  j   AdpRank  i  AND

(6.39)
SInt  j   Thrint AND bat  j   Thrbat

6.4.3 Version 3: AdpSyn
This third version, AdpSpSyn, introduces space popularity to the first version as a factor in
ranking nodes for the purpose of forwarder selection. The ranking function is formulated as
follows:
 Social-Aware Opportunistic Forwarding  123

AdpSpSyn  i   wOpp  Opportunistic

 wIntPower  CumInterst  i   Power  i 
 wSocActive  Social Rank  i   Activeness  i  (6.40)
 wPower Aware  Power  i 
 wPlacePopularity  Place Popularity  i 

where wOpp = 1 − (wIntPower + wSocActive + wPowerAware + wPlacePopularity)

and wPlace Popularity = Pop(Location(i,  now))
As for the place popularity of the node, and as per the recommended variations of Space
Syntax metrics studied in [7], it is computed as the normalized value of the popularity of
the location that this node is most frequently detected in. That is expressed in the following
equation:

PlacePopularity  i   Pop  FreqLocation  i   (6.41)

6.4.4 Version 4: AdpSpSynOp
This fourth version, AdpSpSynOp, includes selecting the next forwarder node if both its
SInt(i, msg) exceeds a predefined interest threshold and its remaining power bat(i) exceeds
a predefined battery threshold. This criterion of selection is added to the selection criteria
defined in the third version which is detailed in Section 6.4.3.
Accordingly, a node is selected as the next forwarder if it satisfies the following conditions:

AdpSpSyn  j   AdpSpSyn  i  AND

(6.42)
SInt  j   Thrint AND bat  j   Thrbat

6.5 EVALUATION

Using simulations, this work evaluates the performance of the implemented algorithms in
terms of certain metrics while forwarding the message within a defined simulation period to
the next candidates. The implemented four versions are compared in performance against the
Epidemic forwarding algorithm as a benchmark, the social-aware opportunistic forwarding
algorithm PeopleRank, and the adaptive power-aware social-based routing algorithm SCAR.
The proposed versions are also compared against three other power-aware social-based
opportunistic forwarding algorithms that incorporate some sort of dynamic weighing of the
ranking attributes; namely, the EBubbleRap, PIPeROp, and PISCAROp algorithms.
It is worth mentioning that the implementation of the EBubbleRap algorithm in these sim-
ulations computes the local and the global ranking functions based on defining the respective
community as follows. The local community is defined based on the user’s interest in the
forwarded content, while the global community is defined based on the social popularity
of the user measured by the number of contacts stored in the user’s device. In addition, the
energy- awareness in the EBubbleRap algorithm is calculated as per the equation defined in
their research paper [14].
124  Opportunistic Networks

6.5.1 Simulation Environment
In order to be able to simulate the performance of the adaptive ranking algorithms within a
dynamic environment efficiently, this research uses the SAROS simulator [10]. The SAROS
simulator is empowered by a set of features and data sets that enable it to simulate social-
aware opportunistic forwarding in mobile opportunistic networks. SAROS imports real
traces from opportunistic network environments such as university campuses, conferences,
and malls, in addition to simulated mobility models such as the SLAW mobility model [26]
for human walks in similar environments. From another aspect, for an accurate simulation
and for logging of the energy consumption of the mobile devices during the forwarding
process, SAROS implements accurate simulation of the power consumption of the mobile
devices simulated in the experiments where four mobile brands are accurately simulated in
terms of their power consumption per activity, and also implements various usage profiles of
these mobile devices. To integrate the social-awareness in the forwarding algorithms, there is
a need to simulate the social relationships among the users moving in the simulated environ-
ment. For that reason, the SAROS simulator imports several forms of social networks and
social profiles. Table 6.1 lists the parameters of the simulation experiment.
The simulations conducted with the SAROS simulator import the mobility traces taken
from the American University in Cairo during Fall 2012. The wireless connection traces of
1760 users were collected from 605 Wi-Fi spots located indoors across campus. Figure 6.1

Table 6.1  Simulation environment parameters.

Parameter Default Value

TTL(msg) 1 hour
Wi-Fi range 50 m
damping factor 0.13 [15]
Thrbat 0.5
Thrint 0.5
Reward 0.5
wcdc 0.25 [6]
wcol 0.25 [6]
Initial Battery Distribution Full battery
Interest Distribution Uniform Discrete Distribution
SInt(destination, message) 0.8–1
SInt(interested node, message) 0.5–0.79
SInt(uninterested node, message) 0–0.49
Mobile usage profiles Suspend, Casual, Regular, Business, and Portable Media
Device [27]
Percent(Destination nodes) 18% of the sample
Percent(Interested nodes) 36% of the sample
Percent(Uninterested nodes) 46% of the sample
Mobile brands Samsung i900, HTC Diamond 2, Galaxyi7500, Spica [27]
No. of WiFi Access points 605 Access Points
No. of Traced users 1760 users
No. of Users with social profile 100 users
Source of traces AUC Campus wireless traces
Duration of simulation 1 hour
Battery model Kinetic Battery Model [28]
 Social-Aware Opportunistic Forwarding  125

(a)

(b)

Figure 6.1  User density heatmap of the AUC wireless traces.

demonstrates a heatmap of a sample of the user density across hours in some high-density
working days during the semester. Figure 6.2 illustrates how overlaying the locations of the
Wi-Fi spots on the university campus map and the floor maps facilitated the computation of
the Space Syntax metrics utilized in the ranking function of the proposed algorithms. Finally,
social profiles of 100 users were collected based on consent from these users.
For the sake of the simulation runs held for this research, the simulator classifies the users
based on their interest in the forwarded content as per a discrete uniform interest distribu-
tion into three categories. The sets are the destination set, in which the users’ interest match
that of the forwarded content with a Jaccard similarity of 0.8 and above, while the interested

(a) (b) (c)

Figure 6.2  Space Syntax graph overlayed on the campus map.

126  Opportunistic Networks

forwarders set includes users with similarity interest between 0.79 and 0.5, and the third set
is the set of uninterested users with similarity interest below 0.5. Different interest distribu-
tions have been examined, but their simulation results are not shared in this chapter due to
space limitation.
As for the initial distribution of the battery level of the simulated nodes, the simulator
starts the runs with full battery levels for all nodes in order to extract the pure effect of each
algorithm on consuming the nodes’ power. Different battery distributions have been exam-
ined, but their simulation results are not shared in this chapter due to space limitation.

6.5.2 Evaluation Metrics
To evaluate the performance of the proposed adaptive algorithms, we apply the following
categories of evaluation metrics: effectiveness, efficiency, and power-awareness.

6.5.2.1 Effectiveness
The algorithm effectiveness is measured by classifying the contacted nodes as per their inter-
est and by measuring the algorithm’s f-measure.

Interest-based effectiveness: An algorithm is effective if it contacts a high portion of the

interested users while simultaneously avoiding the uninterested ones. Our simulator
measures this effectiveness in terms of the ratio of contacted users classified by their
interest. Users are classified as either interested forwarders, destination nodes, or unin-
terested forwarders.
F-measure: Effectiveness is also measured through f-measure as the harmonic mean of
recall and precision [29]. It should be noted here that the targeted true set consists of
the interested forwarder nodes in addition to the destination nodes, while the false
set consists of the uninterested nodes. The three metrics according to their definitions
detailed in another research work [29], we compute to the following equations:

TP
Recall  (6.43)
TP  FN

TP
Precision  (6.44)
TP  FP

Recall  Precision  2
F  measure  (6.45)
Recall  Precision

where

TP = True Positive = all contacted destination nodes and all contacted interested forwarders.
FP = False Positive = all non-contacted destination nodes and all non-contacted interested
forwarders.
TN = True Negative = all non-contacted uninterested nodes.
FN = False Negative = all contacted uninterested nodes.

6.5.2.2 Efficiency
The efficiency of an algorithm is measured in terms of the cost it incurs, the delivery ratio it
achieves, and the delay in time that occurs during the delivery process.
 Social-Aware Opportunistic Forwarding  127

Cost is measured by the count of forwarded message replicas that were generated to
accomplish this process. We measure the total number of message replicas that have
been generated at any given time, and also measure the cost per unit delivery ratio.
Delivery ratio is measured by the portion of successfully reached destination nodes at the
end of the simulation run.
Delay is measured as the average amount of time consumed starting from sending each
message until it reaches one of the destination nodes. Delay also reflects the degree of
user satisfaction. Thus, user satisfaction may be measured by the average delay con-
sumed until a message is delivered to any destination node. The shorter the delay, the
higher the degree of user satisfaction.

6.5.2.3 Power Awareness
To measure the level of power awareness of an algorithm, the total consumed power is
computed and the degree of fairness in utilizing the nodes’ resources is measured.

Power consumption: Algorithm power-efficiency is represented by its ability to conserve

the overall power consumption. This metric is measured by computing the total amount
of power consumed from all of the nodes’ batteries over time. It is also measured by
the total amount of power consumed per unit delivery ratio. In addition, a measure of
the power consumption due to the exchange of control messages is illustrated to evalu-
ate the overload of exchanging control messages and interest information. The power
consumption of control message exchange is represented in the form of a percentage of
the total amount of power consumption.
Utilization fairness: A fair algorithm would not exhaust some nodes’ batteries in message
forwarding while preserving other nodes’ power. That is, it seeks to reduce variance
among the nodes’ battery levels. This is measured through the fairness index of the
algorithm.
Borrowing from the fairness index defined in another research work [30], the standard
deviation (SD) among the simulated population distribution of batteries’ level is divided
by their mean value, then subtracted from one. Equation 6.46 represents the calculation
of the Fairness Index. The index ranges from 0 to 1, where 1 indicates the highest level
of fairness when the SD of the final battery distribution reaches 0.

FairnessIndex  1 

SD Bat  i  | i  all nodes  (6.46)

mean Bat  i  | i  all nodes 
Overhead of Exchanged Control Messages: Part of the power consumed is exhorted as an
overhead of extra control messages exchanged among nodes. This metric is measured in
terms of the number of control messages exchanged among nodes during the forward-
ing process. It is also measured in terms of the percentage of power consumed due to
forwarding these control messages among nodes.

6.5.2.4 Normalized Performance Indices

For a collective performance analysis of the algorithms across various environments, three
normalized performance indices are devised by this research; namely, Effectiveness Index,
Efficiency Index, and Power-Awareness Index. Each of these indices is computed as the har-
monic mean of a group of the aforementioned metrics after normalizing their values. These
128  Opportunistic Networks

indices rely on the harmonic mean instead of the arithmetic mean, since the harmonic mean
of a list of numbers tends strongly toward the least elements of the list, and it tends (com-
pared to the arithmetic mean) to mitigate the impact of large outliers and to aggravate the
impact of small ones [31].
Let us have a look at the detailed calculation of the indices:

Effectiveness Index: This performance index measures the algorithm’s effectiveness

through the harmonic mean of the f-measure, the ratio of the contacted interested
forwarders, and the ratio of the contacted uninterested nodes. Its formula is

 Contacted Interested Forwarders 

 Fmeasure, Total number of Interested Forwarders , 
Effectiveness Index  Harmonic Mean   (6.47)
 Contacted Uninterested Nodes 
 
 Total number of Uninterested Nodes 

Efficiency Index: This performance index measures the algorithm’s efficiency in delivery
through the harmonic mean of the delivery ratio, the normalized value of the paid cost,
and the normalized value of the delay in delivering the message. Since the lower the
cost, the higher the efficiency, the formula utilized subtracts the normalized cost from 1.
For the same reason, the normalized delay is subtracted from 1. The formula is:

EfficiencyIndex  HarmonicMean(Delivery,
(6.48)
1  Normalize Cost  , 1  Normalize  Delay )

Power-Awareness Index: This performance index measures the algorithm’s power aware-
ness through the harmonic mean of the normalized value of the total ratio of consumed
power and the Fairness Index. The normalized value of the power consumption is sub-
tracted from 1, since the lower the power consumption, the higher the power awareness
of the algorithm. The formula is:

PowerAwarenessIndex  HarmonicMean(Fairness Index,

(6.49)
1  Normalize  PowerConsumption )

6.6 RESULTS

This section demonstrates the results of the simulations that are held by the SAROS simula-
tor comparing the performance of the proposed adaptive ranking algorithm versions and the
implemented state-of-the-art social and power-aware opportunistic forwarding algorithms
that cater for some sort of dynamic weighted ranking attributes.

6.6.1 Interest Awareness
When compared to the simulated algorithms, the adaptive ranking versions show improve-
ment in terms of effectiveness and efficiency as follows.

6.6.1.1 Effectiveness
From the Interest-based effectiveness figure (Figure 6.3a), it is evident that effectiveness of
the adaptive ranking versions approaches the effectiveness of the Epidemic benchmark in
 Social-Aware Opportunistic Forwarding  129

(a)

(b)
Figure 6.3  Effectiveness of the adaptive algorithms.

terms of the delivery ratio and the ratio of contacted interested forwarders (except for the
third version, AdpSpSyn, which is very conservative in selecting forwarders). The fourth
version, AdpSpSynOp, significantly reduces the ratio of the contacted uninterested users in
comparison to the other algorithms by about 96%, except for the SCAR versions that totally
avoid them but on the account of attaining a very low delivery ratio. Moreover, Space Syntax
integration within the ranking function (AdpSpSyn and AdpSpSynOp) maintains the same
delivery ratio as that achieved without Space Syntax (Adp and AdpOp) but, crucially, avoids
contact with 86% of the uninterested nodes. It is also noticeable that integrating the inter-
est-threshold opportunistic selection into the ranking function (AdpOp and AdpSpSynOp)
enables contact with a higher ratio of interested forwarder nodes. Compared to the PIPeROp
algorithm, the fourth version, AdpSpSynOp, achieves a higher delivery ratio and a slight
increase in the ratio of the contacted interested nodes and a reduction in the ratio of the
contacted uninterested nodes.
130  Opportunistic Networks

From the f-measure figure (Figure 6.3(b)), it is clear that the adaptive algorithm versions
achieve the highest level of f-measure (except for the third version, AdpSpSyn) where the
f-measure of the fourth version, AdpSpSynOp, is 32% higher than that of the PISCAROp
algorithm and 15% more than that of the PIPeROp algorithm. However, the f-measure of
the non-opportunistic third version, AdpSpSyn, is much less than the opportunistic fourth
version, AdpSpSynOp, because it contacts a lower ratio of interested forwarders. The oppor-
tunistic version, AdpSpSynOp, overcomes this defect through explicit selection of interested
forwarders.
From the F-measure sub-figure in Figure 6.3(b), it is clear that integrating the PI-SOFA
framework with any of the social-aware forwarding algorithm boosts its f-measure such as
the PISCAROp algorithm whose f-measure exceeds the SCAR algorithm by 475.5%.

6.6.1.2 Efficiency
Efficiency is speculated here in terms of the paid cost per unit delivery ratio. If the algorithms
are compared in terms of cost without paying attention to the ratio of gained benefit, a false
conclusion will be reached. Thus, it is much more accurate to study the cost per unit delivery
ratio – as shown in Figure 6.4(a).
The third version, AdpSpSyn, is the most cost-efficient algorithm for exerting the lowest
cost to attain delivery ratio approaching that of the benchmark with an achievement of
75.6% reduction in cost. Although the SCAR algorithm exerts the least cost, its delivery
ratio is the lowest compared to all the other algorithms. Moreover, all four versions of the
adaptive algorithm achieve higher delivery ratio when compared to the PIPeROp algorithm,
yet only the Space Syntax versions, AdpSpSyn and AdpSpSynOp, preserve some cost, ranging
from 13% to 48% reduction in cost. It is also worth noting that by introducing the threshold
opportunistic component to the adaptive ranking with an additional cost, the fourth version,
AdpSpSynOp, successfully approaches the full delivery ratio.
It is noticeable that the adaptive versions exert a small delay in time while the other
PI-SOFA versions incur a long delay in delivery time; thus, the adaptive versions achieve a
93% reduction in delay. It is also worth noting that the fourth version, AdpSpSynOp, incurs
reasonable delay in order to achieve the highest delivery ratio and the highest ratio of con-
tacted interested forwarders out of the four adaptive versions. Overall, the second version,
AdpOp, incurs the least delay among the four adaptive versions. Thus, the adaptive versions
are able to accomplish the difficult task of achieving a very high delivery ratio with a rela-
tively short delay in delivery time, thus competing with the Epidemic algorithm in terms of
the short delay in delivery time on one side, and the high delivery ratio on the other. Over
and above, the adaptive versions contact fewer uninterested nodes, something that Epidemic
fails to maintain.

6.6.2 Power Awareness
The results of the simulation demonstrate the improvement in power awareness of the pro-
posed algorithms in comparison to the benchmark and the social-based power-aware oppor-
tunistic forwarding algorithms. Power consumption and utilization fairness are the two
dimensions of metrics measured to demonstrate the power awareness feature of the exam-
ined algorithms. Furthermore, a study of the power consumed due to the exchange of control
messages in each algorithm is evaluated to reveal the exerted power and communication due
to the overhead of the control message exchange.
 Social-Aware Opportunistic Forwarding  131

(a)

(b)

Figure 6.4  Efficiency of the adaptive algorithms.

6.6.2.1 Power Consumption Awareness

It is more comprehensive and justifiable to analyze the algorithms’ performance in terms of
power consumption per unit delivery ratio, as illustrated in Figure 6.5(a). From this figure,
it is clear that with a minor increase in power consumption, the first version, Adp, suc-
ceeds to achieve an exemplary delivery ratio in comparison to the power-and-interest aware
PeopleRank algorithm, PIPeROp. It is also clear that integrating Space Syntax in the ranking
function, AdpSpSyn and AdpSpSynOp, significantly reduces the percentage of consumed
power per unit delivery ratio by 10%, as it facilitates more effective selection of forwarder
nodes that frequent popular places, and thus speeds up the delivery process. On the other
hand, the SCAR algorithm and its power-and-interest aware version, the PISCAROp algo-
rithm, conserve power, but they achieve the lowest delivery ratio among all of the compared
algorithms. Thus, this power conservation does not meet the desired delivery ratio.
132  Opportunistic Networks

(a)

(b)

Figure 6.5  Power awareness of the adaptive algorithms.

6.6.2.2 Power Utilization Fairness

From the power utilization fairness figure (Figure 6.5(b)), it is clear that integrating the inter-
est-based opportunistic selection in the ranking function slightly improves the level of utiliza-
tion fairness as exemplified in the performance of the second version, AdpOp, compared to
that of the other adaptive versions. However, all the adaptive versions achieve a lower level
of fairness than that of the EBubbleRap and the PeopleRank algorithms. It is interestingly
noticeable how the EBubbleRap algorithm and the Epidemic algorithm satisfy the rule of
“equality in injustice amended” by exhausting the power of the majority of the community
nodes to attain fairness.

6.6.2.3 Overhead of Exchanged Control Messages

In order to measure the overhead posed by the proposed adaptive algorithms in terms of
extra control messages exchanged among nodes, this subsection illustrates the number of
control messages exchanged among nodes during the forwarding process. This section also
presents the percentage of power consumed due to sending these control messages among
nodes.
 Social-Aware Opportunistic Forwarding  133

Figure 6.6  Number of produced control messages of the adaptive algorithms.

Figure 6.6 illustrates the number of control messages exchanged among nodes when apply-
ing each of the proposed adaptive ranking algorithms within the simulation, while Figure
6.7 illustrates the percentage of power consumed due to the control messages sent by these
algorithms. From both figures, it is clear that the fourth version, AdpSpSynOp, consumes
power for the highest number of control messages, while its non-opportunistic third version,
AdpSpSyn, consumes the least number of control messages and the least power in forward-
ing them.
It is worth mentioning that the Epidemic algorithm consumes the highest percentage of
exchanged control message power consumption due to sending the highest number of such
messages. In addition, the PISCAROp algorithm consumes the least number of exchanged
control messages, yet it consumes the highest percentage of power compared to all of the
other algorithms – except for the Epidemic algorithm – which makes it not power-efficient.

Figure 6.7  Percentage of consumed power for the control messages of the adaptive algorithms.
134  Opportunistic Networks

6.6.3 Normalized Performance Indices

The performance of the proposed algorithms is evaluated through the three normalized per-
formance indices defined in Section 6.5.2.4. From Figures 6.8 through 6.10, we analyze the
three performance indices – namely, the effectiveness performance index, the efficiency per-
formance index, and the power awareness performance index – of the simulated algorithms
within the defined simulated environment.

6.6.3.1 Effectiveness Performance Index

From the effectiveness performance index figure (Figure 6.8), it is noticeable that the fourth
version, AdpSpSynOp, maintains the highest effectiveness performance index among all of
the compared algorithms with an increase in the index ranging from 19.7% (when compared
to the PIPeROp algorithm) to 75.5% (when compared to the SCAR algorithm). On another
side, the effectiveness performance index of the third version, AdpSpSyn, is the lowest among
all of the four versions due to its deficiency in contacting many of the interested forwarders.

Figure 6.8  Effectiveness performance index.

Figure 6.9  Efficiency performance index.

 Social-Aware Opportunistic Forwarding  135

Figure 6.10  Power awareness performance index.

6.6.3.2 Efficiency Performance Index

From the efficiency performance index figure (Figure 6.9), it is evident that the non-oppor-
tunistic third version, AdpSpSyn, maintains the highest degree of efficiency with an increase
in value ranging from 24% in comparison to the PISCAROp algorithm, up to 1660% when
compared to the PeopleRank algorithm. The third proposed version of the adaptive algo-
rithm, AdpSpSyn, is cost effective as it pays the least cost among the four adaptive versions
to achieve a quite comparable delivery ratio.

6.6.3.3 Power Awareness Performance Index

From the power awareness performance index figure (Figure 6.10), it is clear that the
algorithms that achieve a low delivery ratio are able to preserve power and maintain a high
level of utilization fairness. Thus, the SCAR algorithm achieves the highest power awareness
performance index value as it achieves the lowest delivery ratio among the compared
algorithms. Note that although the third version, AdpSpSyn, scores the second-highest power
awareness performance index value while maintaining a very high delivery ratio, it fails
to contact many of the interested forwarders, which still leads to a preservation in power
consumption. From the power awareness index figure, one deduces that integrating Space
Syntax in the ranking function improves the power awareness of the algorithms.

6.6.4 Eight-Metric Performance Comparison

Figure 6.11 illustrates the collective performance comparison among the four proposed adap-
tive ranking versions. From the figure, we can notice that version 3, AdpSpSyn, effectively
reduces the paid cost and the delay to maintain the same delivery ratio and to avoid contact-
ing uninterested nodes. However, it achieves the lowest f-measure since it fails to contact the
majority of the interested forwarders. On the other hand, version 4 – AdpSpSynOp – saves
some cost in comparison to the first two versions and avoids contacting uninterested nodes
while maintaining the same delivery ratio and the same interested forwarders contact ratio,
and thus, it achieves the highest f-measure. All of the four versions maintain the same level of
power consumption and utilization fairness.
136  Opportunistic Networks

Figure 6.11  Eight-metric performance comparison.

6.7 DISCUSSION AND CONCLUSION

In this chapter, we argue that there are limitations in social-based opportunistic forwarding
that is oblivious to the variation in importance of the nodes’ ranking attributes as context
changes. It may also be unaware of the absence of some context information as the surround-
ing context varies. More specifically, forwarding algorithms can achieve improvement in
performance in terms of effectiveness, efficiency, and power awareness if they leverage knowl-
edge of the current context of the nodes involved in the process. Such context includes the
density of users in the place, the distribution of power capabilities of the nodes in proximity,
the willingness of the nodes in proximity to share their context information, their willingness
to participate in the forwarding process, or the speed of movement of the users which may
affect connectivity. This context information can be used to continuously adapt the weights
of the ranking function’s attributes. Such attributes include the nodes’ power capabilities,
the user’s interest in the forwarded content, the user’s social popularity, their activeness in
engaging with social contacts, the popularity of the places that are frequently visited, and the
popularity of the place the nodes are currently located in.
This research demonstrates that with such usage of adaptation of the ranking function,
a forwarding algorithm can achieve greater effectiveness, efficiency, and power awareness.
We thus propose a framework that injects adaptive ranking into opportunistic forward-
ing algorithms. This framework is illustrated in the proposed four adaptive ranking versions
that vary in integrating the opportunistic threshold and the Space Syntax awareness. They
are compared to many other interest-and-power-aware social-based opportunistic forward-
ing algorithms which provide some sort of dynamic weighing of the ranking of a func-
tion’s attributes. In such comparison, the Epidemic algorithm is selected as a benchmark.
Furthermore, we devise performance indices in order to evaluate the performance of all the
proposed algorithms.
All the simulations and evaluations presented here were conducted using our own
Social-AwaRe Opportunistic forwarding Simulator (SAROS). This simulator provides an
 Social-Aware Opportunistic Forwarding  137

implementation of a set of state-of-the-art social-based opportunistic forwarding algorithms,

and a variety of devised evaluation metrics. One of the strong features of this simulator is
that it is able to simulate users’ interest in forwarded content as well as the power capabili-
ties of their devices. The simulator also imports a variety of real traces, such as the gathered
mobility traces made by the American University in Cairo (AUC) community and accessed
through the AUC wireless connectivity network. The simulation results demonstrate a great
improvement in performance in terms of effectiveness, efficiency, power awareness, and uti-
lization fairness.
As a final note, given that these algorithms have numerous real-life applications, it is
important to point out that the most suitable algorithm can be determined based on the
needs and requirements of the message senders, the nature of the community of users,
the frequency and duration of contact, the users’ willingness to participate, the capabil-
ity of the devices, or any other limitations in terms of cost and duration. On the basis of
this evaluation, we offer a set of implementation-specific recommendations. One set of
useful applications would be the wireless sensor networks where there are power limita-
tions in the sensors used in this type of network. Accordingly, the advisable forwarding
algorithms would be the power-aware, depletion-rate-aware versions of the proposed
algorithms. From another aspect, for applications that care for forwarding content to
interested parties without entirely contacting uninterested ones, the preferable forward-
ing algorithms would be the interest-aware versions of the proposed PI-SOFA algorithms
[11]. From a different aspect, applications that forward content in environments that are
characterized by a dynamically changing context, the adaptive forwarding algorithms
would fit the most in such environments since these algorithms will adapt the node’s
ranking function according to the variations in context. In such environments, the third
version, AdpSpSyn, is cost effective, as it pays the lowest cost and consumes less power
to achieve a quite comparable delivery ratio. However, relying on both interest-awareness
and Space Syntax, as exemplified in the fourth version (AdpSpSynOp), speeds up the
delivery process and increases the f-measure since it relies on the interested nodes who
frequent popular places.

ACKNOWLEDGMENT

The authors would like to thank Dr. Khaled Harras, Carnegie Mellon University Qatar, for
his efforts within the scope of co-supervision of this work within the PhD dissertation of
Soumaia Al Ayyat.

REFERENCES

[1] D. Xu, Y. Li, X. Chen, J. Li, P. Hui, S. Chen, and J. Crowcroft. A Survey of Opportunistic
Offloading. IEEE Communications Surveys Tutorials, 20(3):2198–2236, thirdquarter 2018.
[2] Chiara Boldrini et al. Modelling Social-Aware Forwarding in Opportunistic Networks. In
Performance Evaluation of Computer and Communication Systems, vol. 6821, pp. 141–152.
Springer, 2011.
[3] Y. Zhao and W. Song. Survey on Social-Aware Data Dissemination Over Mobile Wireless
Networks. IEEE Access, 5:6049–6059, 2017.
[4] H. Y. Bohari and Sh. O. Khanna. Energy Efficient Power Aware Routing Algorithm (EEPARA)
For Mobile Ad Hoc Network (MANET). IJERT, 2(7):1562–1572, 2013.
[5] Soumaia Al Ayyat, Sherif G. Aly, and Khaled A. Harras. PIPeR: Impact of Power-Awareness on
Social-Based Opportunistic Advertising. In IEEE WCNC, pp. 3462–3467, April 2014.
138  Opportunistic Networks

[6] B. Pasztor et al. Opportunistic Mobile Sensor Data Collection with SCAR. In IEEE MASS,
pp. 1–12, 2007.
[7] Soumaia Al Ayyat, Sherif G. Aly, and Khaled A. Harras. On Integrating Space Syntax Metrics
with Social-aware Opportunistic Forwarding. In IEEE WCNC, pp. 1–7, 2019.
[8] Amr Al Jarhi, Hend Adel Arafa, Khaled A. Harras, and Sherif G. Aly. Rethinking Opportunistic
Routing Using Space Syntax. In Proceedings of the 6th ACM Workshop on Challenged Networks,
CHANTS '11, pp. 21–26, 2011.
[9] Ahmed Mtibaa and Khaled A. Harras. Exploiting Space Syntax for Deployable Mobile
Opportunistic Networking. In IEEE MASS, pp. 533–541, 2013.
[10] Soumaia Al Ayyat, Sherif G. Aly, and Khaled A. Harras. SAROS: A Social-Aware Opportunistic
Forwarding Simulator. In IEEE WCNC, pp. 1–7, 2016.
[11] Soumaia Al Ayyat, Khaled A. Harras, and Sherif G. Aly. On the Integration of Interest and Power
Awareness in Social-Aware Opportunistic Forwarding Algorithms. Computer Communications,
71, pp. 97–110, 2015.
[12] Radu-Ioan Ciobanu, Daniel Gutierrez Reina, Ciprian Dobre, and Sergio L. Toral. Context-
Adaptive Forwarding in Mobile Opportunistic Networks. Annals of Telecommunications, 73,
pp. 559–575, 2018.
[13] Pan Hui et al. Bubble Rap: Social-Based Forwarding in Delay Tolerant Networks. In 9th ACM
MobiHoc Proceedings, pp. 241–250, 2008.
[14] Cristian Chilipirea, Andreea-Cristina Petre, and Ciprian Dobre. Energy-Aware Social-Based
Routing in Opportunistic Networks. International Journal of Grid and Utility Computing,
1(3/4):1, 2013.
[15] Abderrahmen Mtibaa, Martin May, and Mostafa Ammar. Social Forwarding in Mobile
Opportunistic Networks: A Case of PeopleRank. In Communication and Social Networks, 58,
pp. 387–425. Springer, 2012.
[16] P Costa et al. Socially-Aware Routing for Publish-Subscribe in Delay-Tolerant Mobile Ad Hoc
Networks. IEEE Journal on Selected Areas in Communications, 26(5):748–760, 2008.
[17] Rares Vernica, Michael J. Carey, and Chen Li. Efficient Parallel Set-Similarity Joins Using
Mapreduce. In ACM SIGMOD Proceedings, pp. 495–506, 2010.
[18] Bill Hillier and Julienne Hanson. The Social Logic of Space. Cambridge University Press, 1984.
Cambridge Books Online.
[19] Abderrahmen Mtibaa and Khaled A. Harras. Select & Spray: Towards Deployable Opportunistic
Communication in Large Scale Networks. In Proceedings of the 11th ACM International
Symposium on Mobility Management and Wireless Access, MobiWac ’13, pp. 1–8. ACM, 2013.
[20] Vassilis Kostakos. Space Syntax and Pervasive Systems. In Geospatial Analysis and Modeling of
Urban Structure and Dynamics. Springer Science, 2009.
[21] Peter C. Dawson. Analysing the Effects of Spatial Configuration on Human Movement and
Social Interaction in Canadian Arctic Communities. In Proceedings of the 4th International
Space Syntax Symposium, 2003.
[22] Bin Jiang. Ranking Spaces for Predicting Human Movement in an Urban Environment.
International Journal of Geographical Information Science, 23(7):823–837, 2009.
[23] Alper Unlu, Ozan O. Ozener, Tolga Ozden, and Erincik Edgu. An Evaluation of Social Interactive
Spaces in a University Building. In Proceedings of the 3rd International Space Syntax Symposium,
2001.
[24] Honglong Chen and Wei Lou. Contact Expectation Based Routing for Delay Tolerant Networks.
Ad Hoc Networks, 36:244–257, 2016.
[25] Chris Chatfield. The Analysis of Time Series: An Introduction. Chapman and Hall/CRC, 6th edi-
tion, 2004.
[26] Kyunghan Lee Kaist et al. SLAW: A Mobility Model for Human Walks. In INFOCOM, pp.
855–863, 2009.
[27] Aaron Carroll and Gernot Heiser. An Analysis of Power Consumption in a Smartphone. In
Proceedings of USENIX, 2010.
[28] J. Manwell and J. McGowan. Extension of the Kinetic Battery. In Model for Wind/Hybrid Power
Systems. Proceedings of EWEC, pp. 284–289, 1994.
 Social-Aware Opportunistic Forwarding  139

[29] David M. W. Powers. Evaluation: From Precision, Recall and F-Measure to ROC, Informedness,
Markedness & Correlation. Journal of Machine Learning Technologies, 2(1):37–63, 2011.
[30] Afra J. Mashhadi et al. Fair Content Dissemination in Participatory DTNs. Ad Hoc Networks,
10(8):1633–1645, 2012.
[31] E. Henry, T.R. Robinson, J.D. Stowe, and A. Cohen. Equity Asset Valuation. CFA Institute
Investment Series. Wiley, 2010.
Chapter 7

Performance Analysis of AODV and DSDV

Routing Protocols in Mobile Ad Hoc
Network Using OMNeT++
Rukhsana Naznin and Md. Sharif Hossen
Comilla University, Cumilla, Bangladesh

CONTENTS

7.1 Introduction and Related Works............................................................................. 141

7.2 MANET Routing Protocols.................................................................................... 142
7.2.1 Proactive Routing Protocol........................................................................... 142
7.2.2 Reactive Routing Protocol............................................................................ 143
7.2.3 Hybrid Routing Protocol.............................................................................. 143
7.3 Destination-Sequenced Distance-Vector Routing (DSDV)....................................... 143
7.3.1 Mechanism of the DSDV Routing Protocol................................................... 143
7.3.2 Advantages of the DSDV Protocol................................................................. 145
7.3.3 Limitations of the DSDV Protocol................................................................. 145
7.4 Ad Hoc On-Demand Distance Vector (AODV)....................................................... 145
7.4.1 Mechanism of the AODV Routing Protocol.................................................. 145
7.4.2 Key Features of AODV.................................................................................. 145
7.4.3 Advantages of the AODV Routing Protocol.................................................. 146
7.4.4 Limitations of the AODV Routing Protocol.................................................. 146
7.5 Simulation Methodology........................................................................................ 146
7.5.1 Mechanism of OMNeT++............................................................................. 147
7.5.2 Simulation Parameters.................................................................................. 148
7.6 Result and Discussion............................................................................................. 148
7.6.1 Performance Metrics..................................................................................... 148
7.6.2 Performance Evaluation................................................................................ 148
7.6.2.1 Impact of Changing the Number of Nodes..................................... 148
7.6.2.2 Impact of Changing the Speed of Nodes......................................... 150
7.6.2.3 Impact of Changing the Packet Length........................................... 151
7.7 Conclusion and Future Work.................................................................................. 153
References....................................................................................................................... 153

7.1 INTRODUCTION AND RELATED WORKS

A mobile ad hoc network (MANET) is a kind of network where communication among

the mobile nodes is established in such a scenario where there exists no fixed infrastructure.
Therefore, the need to establish such a network in the areas of emergency communication,
wildlife tracking, earthquake, terrorist attack, and so on is dominating. This network is a
temporary network where devices communicate with each other with no fixed infrastruc-
ture. The nodes are arbitrary. Therefore, there is no centralized administration due to the
mobility of the nodes in the network. There is an abrupt change of topology. The devices are

141
142  Opportunistic Networks

self-organized and self-configured because of their dynamic nature. They are connected via
wireless links and communicate with each other [1]. To maintain the communication between
two nodes requires some rules which will define the route. These well-defined and specified
rules are called routing protocols. Routing is the network layer task, which accomplishes the
technique of choosing the appropriate path. It will create flooding at the node if broadcast-
ing to all the nodes occurs. So, this should be avoided and also needs an alternative route
as a backup [2]. To create routing decisions, the routing protocol is the fundamental need.
The routing algorithms for wired networks are not suitable for the wireless networks which
are dynamic in nature. The delivery of a packet will not be successful until the topology is
stable; that’s why the changes in topology require a rapid response. A MANET has limited
resources, which is a big challenge for successful communication. So, it requires an efficient
routing protocol that can utilize those limited resources properly and can cope with the rapid
changes in the network under any circumstances, like the size of the network, traffic density,
and mobility of nodes, topology, and even broken routes [3]. The shortest path selection, less
power consumption, and less overhead are the main features for a routing protocol. So, the
routing will be done in minimal time with the maximum utilization of bandwidth, even after
the frequent modification in topology. In this study, we compare the efficiency of routing
protocols by the packet delivery ratio, packet loss, average end-to-end delay, and throughput
by changing the number of nodes, speed, and packet length [4].
Researchers have examined many protocols that are optimum to implement on a defined
network. These researchers have compared performance among some traditional routing
protocols such as AODV, CBRP, PAODV, DSDV, and DSR [5,6]. In [7], the authors conducted
an analysis between AODV and DSDV on the basis of QoS parameter using NS-2. In [8], the
authors review eight routing protocols of MANET, including AODV, DSR, TORA, SSR, ABR,
DSDV, CGSR, and WRP, using NS-2. In this chapter, the performance of AODV and DSDV
is analyzed using OMNeT++.
The rest of this chapter is organized as follows. Section 7.2 discusses the classification
of the MANET routing protocols. Sections 7.3 and 7.4 illustrate the DSDV and the AODV
routing protocols, respectively. Section 7.5 represents the discussion of the simulation tool.
The result and discussion parts are included in Section 7.6, while Section 7.7 includes the
concluding remark with future work.

7.2 MANET ROUTING PROTOCOLS

In MANET, maintaining connectivity is very tough because the topology frequently changes.
Therefore, it requires a stronger and more flexible mechanism to maintain routes. Many
protocols have been introduced for MANETs. These protocols can be classified into three
categories: proactive, reactive, and hybrid, as shown in Figure 7.1.

7.2.1 Proactive Routing Protocol

In this technique, one or more routing tables are maintained by each node in the network
that is updated regularly. Each node sends a printed message to the entire network if there
is a change within the network. This incurs extra overhead cost because it keeps up-to-date
information, which can affect the throughput of the network, but it presents the actual facts
to the availability of the network [10]. In this method, all the nodes are involved in creating a
route. Such techniques react to topology changes, even if the changes do not affect the traffic.
This technique is also called table-driven strategies. Some protocols of this category are WRP,
OLSR, and DSDV [11].
 Performance Analysis of AODV and DSDV  143

Figure 7.1  The classification of MANET routing protocols [9].

7.2.2 Reactive Routing Protocol

Reactive methods have supported the demand for data transmission and are suitable for
ad hoc networks. These types of methods are also known as on-demand methods. In this
method, the routing overhead will reduce when the traffic is minimal. Therefore, the changes
in the topology are not significant since the router’s table is not updated frequently. They find
and maintain a route on which there is no traffic.

7.2.3 Hybrid Routing Protocol

Hybrid routing is the amalgamation of proactive and reactive techniques. This routing
approach shows good performance by using the advantages of proactive and reactive mecha-
nisms. The most common hybrid routings are ZRP, ZHLS, SHARP, and NAMP [12].

7.3 DESTINATION-SEQUENCED DISTANCE-VECTOR ROUTING (DSDV)

DSDV is one of the proactive routing techniques. It is an improved version of the Bellman-
Ford algorithm that guarantees a loop-free routing using a routing table.

7.3.1 Mechanism of the DSDV Routing Protocol

The mechanism of DSDV is very simple. Every entry of the table encompasses a serial
number, and when associate nodes send any update message, the sequence number will be
increased. The routing table’s area unit goes up sporadically once the topology of the com-
munity changes, and the area unit is propagated for the length of the network to preserve
regular records at some point in the network. Each DSDV node continues two preserved
regular records at some points in the network for forwarding and advertising of packets.
The active node will send an update packet to its neighbor node if there is any change in the
topology of the network. After receiving the updated packet from the neighboring node, the
information is extracted by the node, and the router table is updated as follows (Figure 7.2).

i. If the current sequence is greater than the previous sequence, then the route consisting
of the greater sequence will be chosen. Also, the previous sequence number will be
omitted.
144  Opportunistic Networks

Figure 7.2  The flow chart of the DSDV routing protocol.

ii. A route with the smallest amount fee is chosen if the new and existing sequence selec-
tion is the same as this route.
iii. From the newly updated routing table, all the metrics will be chosen.
iv. The process is continued until the whole network units are updated. If a least cost
path is detected, it will be maintained and the rest of the path will be omitted.

There are two processes of updating the routing table in the DSDV:

• Full dump: In terms of routing, all the information that a node needs is contained by
full dump packets. As a result, this full dump approach needs Network Protocol Data
Units (NPDUs) when the routing table is full or near full. If there is occasional move-
ment of node in the network, then packets are transmitted occasionally.
 Performance Analysis of AODV and DSDV  145

• Incremental: The incremental packet can choose the best information as they try to get
the update of full dump packets dispatched by the nodes. So, this approach utilizes the
network resources.

7.3.2 Advantages of the DSDV Protocol

DSDV has some important advantages in MANET. Some of these are as follows [7, 13]:

i . Loop free paths are guaranteed in the DSDV protocol.

ii. DSDV reduces the count to an infinity problem.
iii. Using incremental updates, extra traffic can be avoided instead of doing a full dump
update.
iv. DSDV can reduce the size of the routing table since it maintains not all the paths but
only the effective one.
v. DSDV is a very suitable protocol for an ad hoc network with a lower-size network.

7.3.3 Limitations of the DSDV Protocol

DSDV has some limitations, which are given as follows [7, 13]:

i. There is unnecessary advertising of routing information even if no changes occur in

the network topology. It causes the wastage of bandwidth.
ii. Multipath routing ids are not supported by DSDV.
iii. It is difficult to find out the time delay for sharing the information of routes.
iv. The routing table needs to be updated periodically, which reduces the battery lifetime
and also the bandwidth.
v. Each node needs a table for advertising, so for a large network, the table will be large,
which results in high overhead.

7.4 AD HOC ON-DEMAND DISTANCE VECTOR (AODV)

AODV is one of the reactive types of routing protocols in MANETs. As long as the source
requests for a route, the AODV protocol builds routes between the nodes. AODV does not
produce any additional traffic for communication on links; that’s why it is considered an
on-demand algorithm. Also, to attach multicast group members, trees are formed. Sequence
numbers are used in AODV to make sure the route freshness.

7.4.1 Mechanism of the AODV Routing Protocol

The base of the AODV method is hop-to-hop routing. The node which needs to know the
route of any particular destination will send the route request (RREQ). Then, the transitional
nodes will forward the route request. At a similar time, those transitional nodes will produce
a reverse route to the destination [14]. After receiving the request, a node establishes a Route
Reply (RREP), in which a number of hops are used to reach to the destination. Figure 7.3
illustrates the mechanism of AODV routing with RREQ and RREP.

7.4.2 Key Features of AODV

AODV has the following features [7].
146  Opportunistic Networks

Figure 7.3  Mechanism of RREQ and RREP transmission in AODV [13].

i . It supports unicast, broadcast, and multicast communication.

ii. It establishes an on-demand route with a small delay.
iii. Link breakage in an active route is solved in an efficient way.
iv. It uses sequence numbers to trace the accuracy of information.
v. Through the use of sequence numbers, all routes are loop-free.
vi. It uses periodic HELLO messages to trace the neighbors.
vii. A multicast tree connecting cluster members is maintained for a lifespan of the multi-
cast cluster.

7.4.3 Advantages of the AODV Routing Protocol

The advantages of AODV are represented as follows [7, 13]:

i . Due to the reactive nature of AODV, it can maintain changeable behavior of MANETs.
ii. It supports both unicasts and multicasts.

7.4.4 Limitations of the AODV Routing Protocol

There are some limitations to the AODV routing protocol. These include the following [7, 13]:

i. The nodes within the network will find each other’s broadcasts. It is one of the require-
ments of this algorithm.
ii. There is no reuse of routing information. The routing information is obtained on
demand.
iii. Packets will be distorted due to the insider attackers together with route distraction,
and to incursion.
iv. AODV finds a route after the initialization of a flow which results in high delay of a
large network.
v. High overhead on the information measure.

7.5 SIMULATION METHODOLOGY

A simulation is the re-creation of a real-world process during a controlled environment.

It involves creating laws and models to represent the world, then running those models to
discover what happens. Simulation software is predicted on the method of modeling a true
 Performance Analysis of AODV and DSDV  147

phenomenon with a group of mathematical formulas. It is essentially a program that permits

the user to watch an operation through simulation without actually performing that opera-
tion. In this project, OMNeT++ is used as the simulation tool.

7.5.1 Mechanism of OMNeT++
OMNeT++ is an event-driven framework to build and implement the emulated scenario
of mobile ad hoc networks. It is developed using the C++ language. Figure 7.4 shows the
simulation scenario of OMNeT++. The source of the OMNeT++ model usually contains the
following files [15]:

i. The simple module implementations and other codes are included in C++ (.cc and .h)
files.
ii. The message definitions which are to be translated into C++ classes are contained by
message (.msg) files.
i ii. The description of topology and the components are included in NED (.ned) files;
iv. The assignment of parameters of considered model and other settings are included in
Configuration (.ini) files.

The process to turn the source into an executable form is the following, in nutshell [15]:

i . The message compiler translated Message files into C++ (opp_msgc,).

ii. Object form (.o file) are created by compiling C++ sources.
iii. To get an executable or shared library, object files are linked with the simulation ker-
nel and other libraries.

Figure 7.4  Building and running simulation [15].

148  Opportunistic Networks

7.5.2 Simulation Parameters
Some parameters are essential for the simulation. Simulation details are given in Table 7.1.

7.6 RESULT AND DISCUSSION

The simulated results in details with some performance metrics considered in Section 6.1 will
be discussed here.

7.6.1 Performance Metrics

Packet delivery ratio: The total number of packets delivered to the destination divided by
the total number sent from the source is called the packet delivery ratio.
Packet loss: The subtraction of the number of packets sent from the source and the packets
received to the destination is called the packet loss.
Average throughput: The average number of bytes successfully received by the destination
is referred to as average throughput.
Delay: It is defined as the average delay of successfully sending a packet from a sender
node to the target node.

7.6.2 Performance Evaluation
The simulation is done in different scenarios. The result of the simulation is discussed next.

7.6.2.1 Impact of Changing the Number of Nodes

In this section, the simulation of AODV and DSDV is done for varying the number of nodes,
i.e., 10, 20, 30, 40. Here, the speed of mobile node of 25 mps and the packet size of 128 bytes
are considered. Then, the analysis is done in terms of delivery ratio, packet loss, throughput,
and average delay. When the number of nodes in the network increases, the delivery ratio of
both routings also increases, except when the number of nodes = 40 for DSDV. From Figure
7.5, it is clear that AODV exhibits higher packet delivery compared to DSDV.

Table 7.1  Simulation Settings.

Parameter Value

Simulator OMNeT++
Traffic model UDP
Mobile nodes 10, 20, 30, 40
Simulation time 100 s
Node’s speed 25 mps, 30 mps, 35 mps, 40 mps
Message length 128, 256, 512, 1024 bytes
Transmission range 250 m
Simulation area 600 × 600
Routing techniques AODV, DSDV
Transmitter power 1mW
 Performance Analysis of AODV and DSDV  149

Figure 7.5  Delivery vs. mobile nodes.

For varying the mobile nodes, the packet loss of AODV and DSDV decreases except when
the number of nodes = 40, as shown in Figure 7.6. But AODV exhibits the lower packet loss.
With the increase of the number of mobile nodes, the average delay of AODV routing
decreases rapidly. But DSDV shows comparatively lower delay compared to the AODV, as
shown in Figure 7.7.

Figure 7.6  Packet loss vs. mobile nodes.

Figure 7.7  Delay vs. mobile nodes.

150  Opportunistic Networks

For increasing the number of mobile nodes, the throughput of both routings increases,
as viewed in Figure 7.8. However, the throughput of AODV is significantly higher than the
throughput of DSDV routing (Figure 7.8).

7.6.2.2 Impact of Changing the Speed of Nodes

In this investigation, the analysis for varying the speed of nodes on the basis of delivery ratio,
packet loss, average throughput, and delay is discussed. It is considered that the packet length
is 128 bytes and the number of mobile nodes is 25. The simulation is done for the variation
of the speed of the mobile nodes; they are 25 mps, 30 mps, 35 mps, and 40 mps.
For changing the speed of the nodes, the package delivery of both routings decreases over-
all except the speed of the nodes at 30 mps of AODV routing. As shown in Figure 7.9, the
AODV routing exhibits higher delivery ratio compared to DSDV.
For changing the speed of the nodes, the packet loss increases for both the AODV and
DSDV routings. The packet loss in AODV is significantly lower than the DSDV routing as
shown in Figure 7.10.
For changing the speed of the nodes, DSDV exhibits lower delay than AODV as viewed in
Figure 7.11.

Figure 7.8  Average throughput vs. mobile nodes.

Figure 7.9  Delivery ratio vs. the speed of nodes.

 Performance Analysis of AODV and DSDV  151

Figure 7.10  Packet loss vs. the speed of nodes.

Figure 7.11  Delay vs. the speed of nodes.

The throughput decreases with the increase of the speed of the nodes in the case of AODV,
except at 30 mps. But in DSDV, the average throughput decreases for the increase of the speed
of the nodes. Hence, the AODV routing exhibits higher throughput compared to DSDV, as
shown in Figure 7.12.

7.6.2.3 Impact of Changing the Packet Length

The number of mobile nodes (25) and the speed of the nodes (25 mps) are kept fixed. The
simulation is done for varying the packet lengths, i.e., 128, 256, 512, and 1024 bytes. As
shown in Figure 7.13, AODV deserves a higher delivery ratio compared to the DSDV routing.
The amount of packet loss increases in both AODV and DSDV when the packet length
increases, except when packet length is 256 bytes in AODV and packet length is 512 bytes in
DSDV. But DSDV shows higher packet loss compared to AODV (Figure 7.14).
In AODV, the delay increases with the size of packet length, except at 512 bytes. The value
of DSDV also increases as the length of packet length increases. But DSDV shows lower delay
compared to AODV, as shown in Figure 7.15.
With the increase of packet lengths, the throughput of both routing techniques increases.
It is clear from Figure 7.16 that the average throughput of AODV routing is high compared
to the average throughput of DSDV.
152  Opportunistic Networks

Figure 7.12  Average throughput vs. speed.

Figure 7.13  Packet delivery ratio vs. packet length.

Figure 7.14  Packet loss vs. packet length.

From the discussion of Figures 7.5 through 7.16, it is clear that in all cases, i.e., for chang-
ing the number of nodes, the speeds of the nodes, and packet lengths, AODV routing exhibits
higher delivery ratio, lower packet loss, and higher average throughput compared to the
DSDV routing. However, AODV routing shows higher average delays than DSDV. So, it is a
limitation of AODV routing.
 Performance Analysis of AODV and DSDV  153

Figure 7.15  Delay vs. packet length.

Figure 7.16  Average throughput vs. packet length.

7.7 CONCLUSION AND FUTURE WORK

In this chapter, the functionality of the two routing protocols, namely, DSDV and AODV in
MANETs, has been represented elaborately. Here, the performance of these two approaches
has been analyzed for changing the number of nodes, speeds, and packet lengths in terms of
the delivery ratio, packet loss, delays, and throughput using the OMNeT++ simulator. From
the investigated results of the given scenario, it is observed that AODV outperforms DSDV in
terms of delivery ratio, packet loss, and throughput, whereas DSDV shows minimum delays
compared to AODV. The main limitation of AODV is a high delay. In the near future, AODV
routing can be modified to minimize the delays.

REFERENCES

[1] M. Alslaim, H. Alaqel and S. Zaghloul (2014). A Comparative Study of MANET Routing
Protocols. In The Third International Conference on e-Technologies and Network for
Development, Beirut.
[2] C. Jayakody, R. Samarasinghe and S. Kodituwakku. (2013). Routing Protocols in Wireless Mobile
Ad-Hoc Network-A Review. In The SAITM Research Symposium on Engineering, Colombo.
[3] C. Wahi and S. Sonbhadra (2012). Mobile Ad Hoc Network Routing Protocols: A Comparative
Study. International Journal of Ad hoc, Sensor & Ubiquitous Computing, vol. 3, no. 2 pp.
21–31.
154  Opportunistic Networks

[4] M. Khalaf and B. Mohsin. (2019). Analysis and Simulation of Three MANET Routing Protocols:
A Research on AODV, DSR and DSDV Characteristics and Their Performance Evaluation.
Periodicals of Engineering and Natural Sciences, vol. 7, no. 3, pp. 1228–1238.
[5] A. Boukerche. (2004). Performance Evaluation of Routing Protocols for Ad Hoc Wireless.
Mobile Networks and Applications, pp. 333–342.
[6] T. Salam and M. S. Hossen. (2020). Performance Analysis on Homogeneous LEACH and
EAMMH Protocols in Wireless Sensor Network. Wireless Pers Commun, vol. 113, pp. 189–222.
[7] M. Vijayalakshmi, A. Patel and L. Kulkarni. (2010). QoS Parameter Analysis on AODV and
DSDV Protocols in a Wireless Network. Indian Journal of Computer Science and Engineering,
vol. 1, no. 1, pp. 283–294 .
[8] M.S. Hossen. (2019). DTN Routing Protocols on Two Distinct Geographical Regions in an
Opportunistic Network: An Analysis. Wireless Personal Communication, vol. 108, pp. 839–851.
[9] A. Zaballos, A. Vallejo and G. A. A. J. Corral. (2006). Ad-Hoc Routing Performance Study Using
OPNET Modeler. In OPNETWORK ’2006, Washington, DC, pp. 1–6.
[10] S. Vanthana and V. S. J. Prakash (2014). Comparative Study of Proactive and Reactive Ad
Hoc Routing Protocols Using Ns2. 2014 World Congress on Computing and Communication
Technologies, Trichirappalli, pp. 275–279.
[11] Harsimrankau and J. Singh. (2017). Comparison of AODV and DSDV Protocols along with
Improvement of AODV. International Journal of Advanced Research in Computer Engineering
and Technology, vol. 6, no. 7, pp. 976–981.
[12] K. Pandey and A. Swaroop. (2011). A Comprehensive Performance Analysis of Proactive,
Reactive and Hybrid MANETs Routing Protocols. International Journal of Computer Science
Issues, vol. 8, no. 6, pp. 432–441.
[13] F. Al-Dhief, N. Sabri, M. Salim, S. Fouad and S. A. Aljunid. (2018). MANET Routing Protocols
Evaluation: AODV, DSR and DSDV Perspective. MATEC Web of Conferences.
[14] M.D.I. Talukdar and M.D.S. Hossen. (2019). Selecting Mobility Model and Routing Protocol
for Establishing Emergency Communication in a Congested City for Delay-Tolerant Network.
International Journal of Sensor Networks and Data Communications, vol. 8, no. 163, pp. 1–9.
[15] OMNeT++ Discrete Event Simulator. (2020). https://omnetpp.org/
Chapter 8

Message Forwarding and Relay Selection

Strategies in Mobile Opportunistic Networks
Mohd Yaseen Mir and Chih-Lin Hu
National Central University, Taoyuan, Taiwan

CONTENTS

8.1 Introduction ............................................................................................................155

8.2 Message Forwarding/Routing.................................................................................. 156
8.2.1 Background................................................................................................... 156
8.2.2 Classical Routing Protocols........................................................................... 157
8.2.2.1 Epidemic......................................................................................... 157
8.2.2.2 PRoPHET ....................................................................................... 158
8.2.2.3 Spray and Wait............................................................................... 158
8.2.2.4 Bubble Rap..................................................................................... 159
8.3 Encounter- and Contact-Based Routing................................................................... 159
8.3.1 Study on Real-Life Traces.............................................................................. 159
8.3.1.1 Periodicity/Pattern between Node Pairs.......................................... 160
8.3.2 Relay Selection.............................................................................................. 161
8.3.3 Proposed Routing with Relay Selection......................................................... 164
8.3.3.1 Link Definitions on Regular and Sporadic Links............................. 165
8.3.3.2 RSCR: Scheme Design .................................................................... 166
8.3.4 Performance Results...................................................................................... 166
8.3.4.1 Simulation Settings ......................................................................... 166
8.3.4.2 Results and Discussion .................................................................... 167
8.4 Summary and Future Directions.............................................................................. 168
References........................................................................................................................ 170

8.1 INTRODUCTION

Mobile opportunistic networks (MONs) [1] are sparsely connected mobile ad hoc networks
(MANETs), characterized by low node density and lack of long-term and continuous end-
to-end connectivity between mobile devices. Unlike traditional connected-initiated message
delivery in wired or stationary networks, messages are conveyed through relay nodes by
exploiting pair-wise contact opportunities in MONs. Communication by pair-wise contact
patterns is multi-hop. In this way, each relay node is considered as a router that stores mes-
sages until any next contact opportunity appears to forward the messages. This way of multi-
hop communications is named the store-carry-and-forward data delivery model. Note that, in
the literature, several terms, e.g. data and message, forwarding and routing, and scheme and
strategy, are often used interchangeably, so they are used in the following without ambiguity.
To deal with the inconsistent nature of future contacts, message replication is a naive meth-
odology which has been used to maximize the chances of delivering a message to its final

155
156  Opportunistic Networks

destination [2]. However, the heavy cost of message overhead induced by this sort of naive
schemes is unacceptable, particularly in MONs that consist of resource-constrained mobile
nodes with intermittent connectivity. Knowing the behavior of target users in network con-
texts such as the probability of encountering, contact frequency, and inter-contact time can
predict future contact opportunities and identify better relays for forwarding the messages
towards the destinations in MONs.
To increase the cost-effectiveness of message delivery with high successful delivery rate
and low traffic overhead, the design of an appropriate relay selection strategy requires under-
standing habits in nodal movements and encounter patterns, e.g., average duration, inter-
contact time, meeting frequency, periodicity, etc. Due to the diverse nature of MONs, it is
difficult and challenging to study the movement patterns and come up with unique designs of
efficient message forwarding strategies.
Utilizing social relationships among mobile nodes can help in identifying better relays.
Many attributes of social relationships include but are not limited to betweenness, similar-
ity, and tie strength, which can be incorporated into the decision-making process for relay
selection. The similarity of two nodes implies the number of neighbors these nodes have in
common. Betweenness helps in identifying bridging nodes and carrying messages between
two groups. Thus, relay nodes with higher similarity and centrality values have higher prob-
ability to be selected for forwarding a message [3, 4]. Similarly, the tie strength of a node
reflects its connection strength with the destination node. To further enhance relay selection
strategies, recent research like [5] considered frequency and duration of contacts to identify
bridge nodes. In addition, [6] considered a node with multiple encounters as an appropriate
relay to reflect its ability to forward a message.
Selecting the next relay by understanding the properties of movement behavior of differ-
ent nodes has been quite investigated in MONs [7–9]. Relay selection decisions are taken by
comparing per-node metrics to determine the fitness of the node. Some of the metrics include
quality-based, contact-history and frequency, etc. Thus, a relay node with a higher metric
value is selected as the next relay node. Historical contact information with respect to any
destination node, as studied in [10], is used to form the metrics of contact probability, and
a relay node with higher contact probability with the destination node can be treated as the
best relay.
In summary, this chapter is organized as follows. Section 8.2 introduces the classical rout-
ing protocols used in MONs. Section 8.3 discusses encounter-based routing by first discuss-
ing studies based on real traces in Section 8.3.1, and then explains different relay selection
strategies used to design routing solutions in Section 8.3.2. Section 8.3.3 explains our pro-
posed regular and sporadic contact-based routing (RSCR) scheme, and results are discussed
in Section 8.3.4. Finally, Section 8.4 concludes the chapter.

8.2 MESSAGE FORWARDING/ROUTING

Whereas end-to-end data routing paths are hardly maintained in MONs, conventional rout-
ing protocols that adopted the connection-oriented data delivery in MANETs cannot be suit-
able in MONs. This section gives an overview of routing types and then briefly reviews the
classical routing protocols used in MONs.

8.2.1 Background
Routing in MONs is performed by distributing either a single message copy or multiple mes-
sage copies in the network. The former is done by either direct or first-contact delivery. In the
 Message Forwarding and Relay Selection  157

direct-delivery way, the source node itself waits for the destination node and forwards the
message. In the first-contact delivery, any node that encountered the source node will receive
a message from the source node, and then that encountered node will forward the message to
the destination node. However, the single-copy message delivery in either way is not consid-
ered feasible in highly dynamic networks, because no reliable message delivery is guaranteed
against fragile network topology and constrained resources of mobile nodes in MONs.
Most routing protocols in MONs [1] aim to maximize the message delivery rate, minimize
the transmission latency, or minimize the messaging overhead. These performance measures
are specified as follows:

• Delivery rate means the number of messages created by source nodes are successfully
received by the destinations.
• Transmission latency, or called delay time, is the time span from the time which a
message was sent by the source node to the time which the destination receives this
message.
• Overhead is usually defined as the number of replicas of an original message created in
MONs.

To increase the delivery rate, the same messages are replicated multiple times in the network.
Message replication can be either restricted or unrestricted. With a restricted replication-
based routing way, only a fixed amount of message copies are distributed to different nodes.
By contrast, in an unrestricted routing way, no fixed limit is applied to the number of replicas
of a message after the source node created this message. In addition, in order to regulate the
network and its storage resources, each message has a time-to-live (TTL) value. When a mes-
sage expires, it is dropped from the buffer space of a node. To improve the delivery rate, any
routing scheme should attempt to deliver messages to their destinations before their associ-
ated TTL expires.

8.2.2 Classical Routing Protocols

Following the store-carry-and-forward data delivery model in MONs, source nodes attempt
to distribute messages to destinations with the help of relay nodes. Two neighbor nodes
can exchange messages kept in local buffer whenever they are in contact with each other.
A contact is defined as a period when two nodes temporarily stay in a reciprocal transmis-
sion range during their movement in a network. In general, a node may be in contact with
several other nodes at the same time, so messages can be handed over across neighbor nodes
quickly to expedite the message distribution. To provide a background knowledge of design
and development of mobile opportunistic routing protocols, the following briefly mentions
some of the classical routing protocols.

8.2.2.1 Epidemic
One of the earliest mobile opportunistic routing protocols was designed to forward a mes-
sage whenever two nodes contact with each other [2]. Epidemic belongs to the unrestricted
type, since this scheme floods messages into the network by duplicating a message copy to
each encountered node. As long as the encountered node does not keep the same message
copy, a message is forwarded to it as the source node. Any node carrying the message copy
will continue to forward more copies to other encountered nodes if they did not receive that
message before. In this way, a great number of message copies of the original message from
the source node can be broadly distributed in a network. Eventually, the destination node can
158  Opportunistic Networks

have the maximum possibility of receiving the message eventually. Admittedly, it is known
that the Epidemic protocol consumes considerable resources of nodes, such as buffer space
and energy battery, and causes heavy message overhead in a network. To maintain the system
performance, Epidemic can be used fairly subject to the provision of sufficient buffer space
and definite message lifetime, i.e., TTL.

8.2.2.2 PRoPHET
PRoPHET [10] belongs to the unrestricted type of routing. In PRoPHET, a node forwards
a message copy to another node only if the latter node has higher delivery predictability
(DP) than the former node. The value of DP implies the probability of one node meeting
with other nodes in MONs. Each node maintains a table of DP values with different nodes.
Whenever two nodes meet, their corresponding DP values are updated according to Equation
8.1. In Equation 8.1, P(x, y) and P(x, y)old represent current and previous DP of node x with node
y, respectively. The value Pinit ∈ [0, 1] is an initial constant.

 
P x,y   P x,y old  1  P x,y old  Pinit

(8.1)

P x,y   P x,y old *  k (8.2)

 
P x,z   P x,z old  1  P x,z old  P x,y   P x,z   

(8.3)

If x and y do not contact each other for a certain period, their DP values will decay accord-
ingly, and the updated DP value is calculated by Equation 8.2. In Equation 8.2, γ ∈ [0, 1] is
an aging constant, and k denotes the number of time units that have expired since the last
update of this DP value. Equation 8.3 updates a transitive DP value of x with node z where
β is a parameter with value ∈[0, 1].

8.2.2.3 Spray and Wait

Spray and Wait (SNW) [11] belongs to the restricted type of routing where a fixed number of
messages copies are replicated in the network. SNW consists of two phases, the Spray phase
and the Wait phase, as follows.
In the Spray phase, SNW distributes at most L copies of a message to different relay nodes.
In source spraying, a source node itself spreads L message copies in MONs whenever it
comes into contact with other relay nodes. However, if the source node contacts the destina-
tion node in between, the message is directly forwarded, and no message copy is distributed
further. Otherwise, the relay nodes that received the message copy from the source node waits
for the destination node to transfer the message. In binary SNW, a source node initially trans-
fers n = L/2 messages where L > n > 1, copies to the first encountered relay node, and keeps
n = L/2 to itself. Later, both the source node and the encountered node transfer half of the
remaining n/2 copies to other relays, which is the same as the previous iteration. This spray-
ing process continues until each node retains a copy with n = 1 and waits for the destination
node to transfer the message directly.
Provided that the destination node is not encountered during the Spray phase, a relay
node goes into the Wait phase. After that, a relay node carrying a message copy will wait to
send that message directly to the destination node once they come into contact with each
other.
 Message Forwarding and Relay Selection  159

8.2.2.4 Bubble Rap
Bubble Rap [4] exploits social relationships among different nodes to forward a message.
Such social relationships are built by following the interaction among different people,
recorded by using their mobile devices. This interaction helps in predicting the next contact
time between mobile nodes, and thus enhances the delivery rate. Social metrics such as cen-
trality and community based on examination of a node’s past interactions are often jointly
considered for forwarding a message in social-based MONs.

• Local centrality: This relative rank is allocated to a node and is used to forward a mes-
sage within its group/community.
• Global centrality: This rank allocated to a node reflects its ranking in the entire MON.
This rank is utilized across the whole MON for forwarding a message.

Initially, a node with a high global rank is given a larger priority to be a message transfer.
This message is replicated before it reaches a node belonging to the same community as the
destination of the node. Afterward, the local ranking within a community is used to further
disseminate the message until it reaches the destination node.

8.3 ENCOUNTER- AND CONTACT-BASED ROUTING

Encounter-based routing techniques, a.k.a. contact-based routing, deal with contact histories
between mobile devices. Because some of the nodes in MONs may contact frequently, such
nodes may help to increase data delivery rate cost-efficiently rather than other nodes that
do not have apparent contact relationship. For example, [8] uses the past encounter rate as
a metric to decide the number of replicas of a given message which would be sent to other
nodes. Note that the terms encounter and contact are used interchangeably hereinafter, while
they are commonly presented in the literature.
Our goal in this section is to mention recent efforts of encounter-based routing studies in
MONs. Encounter patterns in the context of nodal moment behavior correspond to the man-
ners which nodes meet with each other during movement in MONs. By exploiting encounter
patterns, mobile nodes can be guided to select the next relay for message forwarding. Some
practical mobility features, such as repeated appearances at the same location, repetitive
patterns, and strong correlation between mobile nodes, can be exploited to obtain either
periodicity between node pairs or differentiate sporadic and regular contacts among nodes.
The next subsection describes previous studies on real-life traces and then explains different
routing algorithms designed by reference to various encounter patterns in MONs.

8.3.1 Study on Real-Life Traces

Meeting time between node pairs can be considered a key element to identify the repetitive
contacts in the MONs. Many studies on human mobility have rendered a thorough under-
standing of nodal encounter patterns [12–17]. Various techniques, e.g., auto-correlation
function, Fourier transform, correlation analysis, time-variant, etc., are employed to examine
some measurement results based on wireless and mobile trace data sets.
The sort of trace data sets in wireless and mobile environments include Bluetooth device
proximity, T-mote sensor devices carried by users, and Wi-Fi access points (AP) placed inside
the university campus and at different locations for data collection. The data collected include
the node’s address, associated AP, and the timestamp for start and end contact time between
160  Opportunistic Networks

mobile devices. The processed information on such trace data includes duration, location,
and timestamps of encounters. Some of these traces in the research community can be found
here [18]. In what follows, we will go on to understand mobility patterns in reference to the
studies of [12, 14–16] which reported some measurement results based on these traces.

8.3.1.1 Periodicity/Pattern between Node Pairs

Given a specific time period, a contact between any two nodes occurs frequently is referred
to as a periodic contact. Finding periodicity between node pairs can help in predicting the
next contact time and can identify a relay with a shorter inter-contact time with the destina-
tion node.
The study in [14] investigated the periodicity in encounter patterns using power spectral
analysis. They used the auto correlation function (ACF) to identify repetitive patterns and
also used the discrete Fourier transform (DFT) to find distinct periodicity. Encounter rate was
used as a metric to group encountered pairs and individual nodes to highlight any periodicity
involved in each of them. To analyze the encounter pattern in a trace data, different variables
such as encounter frequency Fd(i,  j), daily encounter Ed(i,  j), hourly encounter Eh(i,  j), and
duration of encounter Ld(i,  j), where i and j represent encountered nodes, are used. ACF is
then applied to this time-domain encounter trace to find the repeated periodical patterns.
ACF measures the correlation between observations at different lags or distances apart. At
the lag of k = 0, the auto-correlation is the maximum, resulting in a variance δ. If λ represents
the mean rate with respect to a pair (i,  j) in a total trace time of T days, the auto-correlation
coefficients for each lag of k(1 < k < T) with respect to (i,  j), which is denoted as rk(i,  j) and
can be calculated by Equation 8.4:


T k
 Ed     Ed k   
rk  i, j   d 0
(8.4)
 d 0  Ed   
T 1 2

The resulting series gives a repetitive pattern for certain lags. DFT is then applied to find
distinct/hidden periodic properties by transforming autocorrelation coefficients of the time
series data to the frequency domain as:

T 1 2i


 kc
yc  i,j   rk  i,j  e T , (8.5)
k 1

where yc(i,  j) is a power spectrum of the pair (i,  j) for each frequency component c(1 ≤ c < T).
To further observe the periodicity of an encounter pair, the average auto-correlation coefficient
of each lag is calculated and then transformed to the frequency domain.
Similarly, to derive periodic patterns, [12] examined skewed and periodic reappearances at
the same places. They observed from different traces and found that, on average, 65% time
is spent by a node with one AP, and more than 95% of online time at as few as five APs. The
finding of periodical re-appearance signifies the re-appearance of a node at the same AP after
some fixed periods. To capture skewed locations, they defined some popular locations for a
node by defining some communities for each node. Thus, a node visits its community more
often than others.
Re-appearance of a group of people in “only a specific period” is known as a transient com-
munity (TC). To detect TCs, [16] proposed a Contact-burst-based Clustering Method (CCM)
by utilizing contact pairs. Each pairwise process is formulated to be a regular presence of
 Message Forwarding and Relay Selection  161

contact bursts, which is defined as a period when contacts regularly occur between two node
pairs. TCs are detected by clustering the node pairs with similar contact bursts. Members of
a node pair may have frequent contacts with each other in the same duration. Thus, a set of
contact bursts with a similar duration will form a TC.
To detect frequently occurring contacts and predict the next contact time between any node
pair, [15] used the correlation analysis method, called “Auto Regression Moving Average.” To
detect a pattern, each node stores a contact history list with a fixed length, including contact
time (up and down), meeting amount to obtain encounter history, contact frequency, and
contact interval. Using these variables, the meeting probability can be calculated to deter-
mine nodes with definite patterns. Correlation analysis is used with derived nodes to obtain a
confidence degree, which is then used to improve prediction accuracy. This process is briefly
described next.
First, a transaction set is characterized as a sequence of events, where an event is a meeting
of two nodes at any time. The support analysis and confidence analysis of the transaction
set, as will be described later, are used to reveal the pattern of a mobile node. For any node
with the encounter history, (Z1, …, Zn) is termed as an event set S. For any two consecutive
encounters (Zi, Zi+1) in S, it is defined as an event. For any recent event, (Zn−a, Zn) denoted A,
its support degree is calculated by Equation 8.6:

N  A
support  A   , (8.6)
N  All2 

where N(All2) represents the occurrence number of all two consecutive encounters event in
S, and N(A) represents the occurrence number of A in S. For each node Y, let B be an event
for two consecutive encounters (Zn, Y). Then, the support degree of event A ⇒ B can be cal-
culated by Equation 8.7:

N  A  B
support  A  B  , (8.7)
N  all3 

where A ∪ B represents the three consecutive encounters event (Zn−1, Zn, Y), N(A ∪ B) repre-
sents the occurrence number of A ∪ B in S, and N(all3) represents the occurrence number of
all three consecutive encounters event (Zn−1, Zn, Y) in S. Confidence degree is then defined to
denote the conditional probability of the event A ⇒ b:

support  A  B
confidence  A  B  . (8.8)
support  A 

To determine the strong associated pattern for the contact pattern A ⇒ B, a minimum
confidence threshold minsup ∈ [0, 1] is used. If confidence(A ⇒ B) > minsup, then A ⇒ B has a
strong association. Here, the Time Series Analysis Model (TSAM) is used to estimate the next
contact time. TSAM is a statistical method for discovering the dynamic patterns in a dynamic
data set. With the obtained strong association, a dynamic time series is extracted, and then
ARMA takes it as an input to predict the next contact time.

8.3.2 Relay Selection
Having established the presence of pattern periodicity between mobile nodes in MONs,
we now focus on some of the forwarding strategies by further exploiting these patterns to
162  Opportunistic Networks

enhance delivery rate cost-effectively. Upon prior studies in the literature [6, 17, 19–23], we
describe some strategies in regard to appropriate relay selection methods in MONs.
Using social network concepts such as centrality and communities, [21] proposed an ana-
lytical model for multicast relay selection to minimize cost in terms of the number of relays
used. The centrality metric measures the ability of a node to communicate with other nodes.
With a single data source and N nodes in a network, this work proposed the cumulative
contact probability (CCP) for a node i with a mean rate λi,j. Then, a centrality metric is
defined in Equation 8.9. A node with a higher CCP implies its effectiveness to deliver data
to destinations.
N


1
Ci  1  e  i , j T . (8.9)
N  1 j 1, j  i

In Equation 8.9, Ci represents the average probability that a node randomly contacts i
within T. Data forwarding to all the destinations will start according to a specific threshold
value: delivery ratio p. To forward data, a minimum number of relays are selected to sat-
isfy both p and T constraints. With multiple data sources, a community-based approach is
utilized to forward data. A node keeps track of nodes that have contacts with destinations
within its social community and selects relays accordingly.
To optimize delay and improve performance, [19] proposed the conditional intermeeting
time (CIMT) metric. Using past encounter history, CIMT calculates the average inter-contact
between two nodes relative to a meeting with a third node. To take forwarding decisions,
CIMT is used to the compare meeting time with the destination node whenever two nodes
contact each other. If i, carrying a message, meets j and learns j’s CIMT is less than itself, it
selects j as the next relay to forward the message.
To improve performances of data forwarding in MONs, [20] exploited the transient social
contact patterns. These patterns, discussed later, are formulated based on experimental stud-
ies on data traces. Further, the centrality of a node is evaluated using the transient social
patterns. In the following, we first describe the transient social patterns, and then data for-
warding and centrality.

• Transient contact distribution: At different periods, the contact distribution is highly

skewed. For example, during the daytime, nodes frequently contact other nodes, in
contrast to the situation at nighttime. Nonetheless, at nighttime, some nodes may still
be able to transfer data using transient contact distributions.
• Transient connectivity: This includes contacts that occur merely in some specific peri-
ods. For example, students in a class remain connected with each other only for that
specific time.
• Transient community structure: A node belongs to different communities during spe-
cific periods. For example, a student at some university may be connected with class-
mates during the daytime and with roommates during the nighttime.

Data forwarding is mentioned below. If node i meets node j, then i will forward the data to j
in the following cases only:

1. If j is the destination node, or both the destination node and j belong to the same
community.
2. If the first condition is false, then data forwarding metrics for both i and j are calculated
by their transient contact patterns. If j has a higher metric value than i, then i forwards
the data.
 Message Forwarding and Relay Selection  163

The centrality value is used as a metric for data forwarding. It measures the expected number
of nodes that a node can meet within the given time. At time tc, the centrality metric Ci of i
is defined by Ci = ∑j ∈ Ncij, where N indicates the set of nodes, and cij implies the number of
nodes expected by i to encounter within some time constrains by contacting j.
The authors of [17] investigated social contact patterns from the temporal contact patterns
in order to predict future contact patterns, and thus improved the cost-efficiency of data for-
warding in MONs. A new metric, called temporal closeness (TCL), is defined to capture the
presence of social contact patterns. Average separation, which includes both the frequency and
duration of contact between two nodes, is utilized to deduce TCL. Let i contact j at time t1, and
let the duration of contact be td1. Now, if both i and j contact again at t2 such that t2 > t1, then
the separation time is equal to t2 − (t1 + td1). A shorter average separation time reveals a closer
relationship, and the variance of the separating time reflects irregularity. To derive a temporal
centrality metric value, the following procedure is adopted.
Let θuv and δuv respectively denote average separation (AVuv) and variance of separation
time (VAuv) between two nodes u and v in a certain window (w). The exponential function is
used to normalize both the average separation time and average separation in each time win-
 uv  uv
1 1
dow. Then, given with AVuv  e w and VA uv  e w , TCL between u and v is measured as
TCL(uv) = g × AVuv + (1 − g) × VAuv, where g ∈ [0, 1]. Finally, taking all V nodes in the MONs
into consideration, the temporal centrality (TC) in N time windows is used as a centrality
metric based on TCL of pairwise nodes, expressed in Equation 8.10.


1
TC  u   TCL  uv  . (8.10)
N  1 vV ,v  u

The Recent Weighted Method (RWM) is proven to be best for predicting future TC. Herein,
we skip the description of the prediction method; readers are encouraged to see [17] for fur-
ther details. With some time constraints on the destination node, a node with a higher TCL
is selected as the next relay node. Note that if there is no relationship with the destination
node, then the future TC is utilized.
Using the concept of social energy metric, [6] proposed a Social Energy Based Routing
(SEBAR) protocol. A node with multiple encounters will have a higher social energy and
can be considered a better relay. SEBAR uses the concept of community to calculate social
energy. Let Nk, l be the amount of energy generated when nodes k and l encounter each
N
other. If k belongs to one community, then it contributes p × k,l to its community and
2
N
keeps 1  p   k,l , where p represents percentage. Else, if it belongs to nk communities, then
2
N k, l
p× is shared with each community. By this process, the community gains the energy
2nk
from its members and distributes this energy to its members based on the node’s community
centrality, which is defined in Equation 8.11.


mk
Dk  i 
cj  j   i 1
. (8.11)
 
mk
Dk  i 
any vx C j i 1
In Equation 8.11, ck(j) represents the community centrality of k in community Cj and with
mk

a total number of mk encounters of k, D i  is the total contact duration, where D (i)

i 1
k k
164  Opportunistic Networks

mx

represents the duration of the ith contact. Similarly,  D i  represents the total con-
any vx C j i 1
x

tact duration of all member nodes in Ck. The energy   k  gained by k consists of two parts:
the reserved energy during collision  E_N k  and the gained energy from communities  E_Ck ,
as calculated in Equation 8.12.
mk nk

Ek  i 1
E _ Nk  i   E _ C  j  ,
i 1
k (8.12)

N k, l  i 
E _ N k  i   1  p   ,
2

N x, l  i 
mx

E _ Ck  j   ck  j    p  2n x
.
any vx C j i 1

However, if a node does not encounter other nodes for a long time, its energy decays. An
energy decay coefficient is defined by ϵ(0 ≤ ϵ ≤ 1), so that after each τ, the energy of a node
will decay by ϵ. So, Equation 8.12 is defined again and rewritten in Equation 8.13.

Etk  E _ N tk  E _ Ctk , (8.13)

mkt

E _ N tk  1     E _ N tk   E _ N i  ,
i 1
k

 mtx
N x,l  i  
E _ C  1     E _ C
t
k
t 
k    ck  j  

any C j k 
 p
2nx 
,
any vx C j i 1 

where tk, E_N tk , and E_Ctk denote the energies gained at time t, and mxt represent a number
of encounters of node x in the period [t − τ, t]. Now, consider k carries q message copies for
the destination d and encounters l. Node k forwards half of the copies in two cases: (1) k
does not belong to d’s community, and k < l; (2) both k and l belong to d’s community and
E_Ck < E_Cl .
In [22], the design of selecting the next relay for data forwarding with the shorter inter-
contact time to the destination was considered. A utility value for each relay node, based on
the features such as periodicity and similarity, was designed to select the next relay.
The authors of [23] developed a new comprehensive metric to build an accurate social
relationship between different nodes in a social graph. In particular, they considered several
attributes, including average contact duration, shortest separation time, contact frequency,
and average separation time. Two nodes with frequent connections create stronger relation-
ships, which means that the weight, a.k.a. strength, between them is higher. Using this new
metric, they also designed a method to detect the hierarchical community structure. Finally,
by exploiting the overlapping community structure, they proposed a routing strategy in
social-based MONs.

8.3.3 Proposed Routing with Relay Selection

Our study efforts specify a novel decision-making process for contact-based relay selection
in MONs. To explore the use of contact patterns to increase delivery rate cost-effectively, we
examine the intrinsic properties of contact patterns and clarify two contact types, i.e., regular
 Message Forwarding and Relay Selection  165

and sporadic contacts. Contacts occurring periodically are defined as regular, and occasional
contacts are termed as sporadic. Readers can refer to [24] for further details. The content
that follows first explains the finding and differentiation of these contact patterns in Section
8.3.3.1. In Section 8.3.3.2, we describe the proposed routing scheme, named Regular and
Sporadic Contact-Based Routing (RSCR).

8.3.3.1 Link Definitions on Regular and Sporadic Links

To discover contact patterns in a time scale, we first intend to determine the link type cor-
responding to the contact behavior. Let a link define a contact between two nodes nx and
ny, expressed as Lx ⇌ y where Lx ← y indicates ny contacts nx. To be a regular link, we say that
Lx ← y is a regular link Rx ← y if nx and ny contact each other multiple times over a particular
threshold value in MONs. Oppositely, a sporadic link Rsp(x ← y) is formed when ny contacts nx
only sometimes in a random time period Tr. If there exists a regular link between nz and ny,
then a link between nx and ny is a transient link and is defined as

 
Ltrx  y  z | if  Rx  y | Ry  z  .

We now plainly use the mean inter-contact time as a specific time period to define any
occurrences of either regular or sporadic patterns by considering the following two cases:

1. If nx contacts ny at times t0, t1, t2, …, tM, then we obtain M inter-contact time samples
as {(T = t1 − t0), (T2 = t2 − t1), …, (TM = tM − tM − 1)}. Then, the mean of inter-contact time


M
Ti
is calculated as T x  y  i 1
. Node nx can contact other nodes, and we can obtain
M
different means of inter-contact time of nx. Given {Tx←1, Tx←2, ……, Tx←Z} with respect to
Z distinct nodes encountered by nx during M time periods, we will average these means
to be


Z
T xy
x y 1
T  . (8.14)
Z

2. In this case, we update Tx←y incrementally with TMx← s

y
in the previous number of Ms
samples and the current sample TMs + 1, i.e. tMs + 1 − tMs. Thus, Tx ←y can be updated as

Ms  TMx y
 TMx y
s 1
T xy  s
. (8.15)
Ms  1
Thus, Tx/Tx←y is used as the specific time period to distinguish regular and sporadic links
associated with nx.
With the link type definitions and mean of inter-contact time, we briefly describe how to use
Tx/Tx ← y to extract any link types between two nodes. We consider Lx ← y as an initial state (s0) in
a Markov chain. The change of a state occurs whenever nx or ny meets another node to form a
new link and form a new state. If nx and ny again contact each other in the near future, we call
to s0. If fLx  y Lx  y denotes the first recurrence to Lx ← y at the nth step,
n
it a recurrence fL
x ← yLx ← y

then its probability is defined as fLx  y Lx  y  P sn  Lx  y | s0  Lx  y , s1  Lx  y , , sn 1  Lx  y ,

n

where sn denotes the nth state. The probability of recurrence to Lx ← y can be written as
166  Opportunistic Networks

FLx  y  f  
n 1
n
Lx  y Lx  y .

(8.16)

Then, s0 is a recurrent state, if FL = 1. Thus, Lx ← y is regular if there is at least one recur-

x←y
rent state in Tx or Tx ← y. Otherwise, Lx ← y is sporadic, and may occur at Tr.

8.3.3.2 RSCR: Scheme Design

The RSCR scheme is divided into two parts:

• Link type determination: Given that each node records contacts encountered in time t,
the link determination process determines the link type between nx and ny by Tx/Tx ← y
together with the fixed time length T. Given that T < t, T is used to get a certain number
T
of time periods with respect to Tx by x . Then, if a Lx ← y i.e., FLx ← y = 1 is formed within
T
T
t in each Tx, Lx ← y is regular. If Tx > T, x is set as 1 by default. Later, Tx ← y in Equation
T
8.15 is used to decide the link type for Lx ← y formed after t. Let nx contact ny at time t1,
t2, and t3, such that t1 > t and t2 > t1. The first two inter-contact samples are calculated as
 
T1x  y  t1   t0  0  , and T2x  y  T1x  y   t2  t1  / 2. Until now, Lx ← y is considered as
sporadic. Now, If T2x ← y is greater than t3 − t2 and t3 − t1 is less than T, Lx ← y is taken as
regular. If this proposition is false, a link is sporadic and the determination process stops
here. Otherwise, the same steps are run to check the new contacts between nx and ny.
• Message forwarding policy: To select the next relay node to forward a message, let nx
contain a set of messages {mx} in local buffer for the destination node nd. As with con-
tacting ny, nx applies the RSCR scheme to take a decision by considering the link type
of Lx ← y, a threshold α > 0, and the remaining TTL value of every mx in {mx}, denoted as
TTLxR . Node nx takes a decision to select a next message in mx to send it to ny by refer-
ring to the following cases:
• If either ny is the destination node or Lx ← y is Rsp(x ← y), Ry ← d, or Ltrx ← y ←d , nx forwards
mx to ny.
TTLx
• If the value of x ← yR is smaller than α, nx forwards mx to ny without considering
TMs
Lx ← y. Note that when Lx ← y is updated either regular or sporadic, this condition is
not followed further.
TTLxR
• If Lx ← y is Rx ← y and the value of x is greater than α, nx retains mx. Otherwise,
it forwards m . T | T x←y
x

8.3.4 Performance Results
In this section, we present the simulation settings and compare the performances of RSCR
with Epidemic [2] and PRoPHET v2 [10].

8.3.4.1 Simulation Settings
We use the opportunistic network simulator ONE [25] with two mobility benchmarks, i.e.,
the SLAW [13] mobility model and real trace-based dataset Infocom ’05 [26]. FIFO dropping
policy is adopted in all the schemes to drop a message whenever a node’s buffer operates at
its limit. Table 8.1 lists the simulation parameters.
Following two performance metrics are used.
 Message Forwarding and Relay Selection  167

Table 8.1  Simulation parameters.

Parameters Infocom ’05 Slaw

Number of nodes 41 100

Map size – 2000 × 2000 m2
Simulation time 24 hours 12 hours
Contact record collection time (t) 12, 24 hours 9, 12 hours
Time-to-live (TTL) duration 1 to 10 hours 1 to 5 hours
Message size 50 KB 50 KB
Buffer size 20 MB 10 MB
Transmission radius – 10 m
Number of sources 1 1
Message creation intervals 100 seconds 50 seconds
Transmission speed 2 MB/s 2 MB/s

• Delivery rate: The ratio of the number of messages delivered successfully to the total
distinct messages created by all source nodes.
• Overhead ratio: The ratio of number of messages relayed between relay nodes to the
distinct messages created by all source nodes.

8.3.4.2 Results and Discussion

This section describes the practical usage of α and T, and then relative performance among
RSCR, PRoPHET v2, and Epidemic.
Experiential values of α and T are gathered from experimental trails according to the mean
inter-contact time and number of nodes in the network. In order to decide appropriate values
of α and T to achieve good performance of RSCR, we intend to find fit values using SLAW
TTLxR
and Infocom ’05. According to the comparison between x and α, a larger α means a
T | T x←y
message will be forwarded earlier, and a smaller means keeping a message in a buffer for a
T 
longer time. Similarly, a larger T implies the number of time periods by  x  comes out more
T 
as compared with a smaller T. In general, the variance of the following factors can influence
the performance:

1. Number of regular and sporadic links: RSCR greedily forwards messages when
sporadic links are encountered, which results in an increase in overhead. With more
time periods, the condition for a regular link is not true for each time period, which
results in more sporadic links. With a smaller T, more regular links are produced.
As shown in Figure 8.1, the cases as T = 3 hours cause more overhead than those as
T = 0.5 hours.
2. Node density: In Infocom ’05 with low node strength, we take a large value of α to
disseminate messages. SLAW has higher node density, so a smaller value of α is used to
wait for a better relay in the near future.
3. Time length of TTL: With a sufficient buffer size, a message with a longer TTL value
has more chances of reaching the destination. TTLxR also becomes one of the dominat-
ing factors in the message forwarding decision.
168  Opportunistic Networks

200 α = 0.05 200 α = 0.05

175 α = 0.1 175 α = 0.1
150 α = 0.2 150 α = 0.2

Overhead
Overhead

125 α = 0.3 125 α = 0.3

100 α = 0.4 100 α = 0.4
75 75
50 50
25 25
0 0
1 1.5 2 2.5 3 3.5 4 4.5 5 1 1.5 2 2.5 3 3.5 4 4.5 5
(a) Time to Live (Hours) (b) Time to Live (Hours)

Figure 8.1  Overhead ratio by RSCR with different α, T, and TTL under SLAW (baseline: (a) T = 30, and (b)
T = 180).

Performance results are shown by varying TTL = [1,5] and [1,10] hours in SLAW and
Infocom ’05, respectively. RSCR is implemented by considering three variants, i.e., Cases 1,
2 and 3, as described next.

• Case 1 – RSCRbase: We calculate the mean inter-contact and link type information
between each node pair in a network in advance. To measure the performance of RSCR,
we vary TTL from 1 to 5 hours and use T = 1 hour. Epidemic and PRoPHET v2 are
sensitive to number of nodes in a network, encounter frequency, and message genera-
tion. With particular values of α and T, a longer TTL makes RSCR to keep messages
longer on nodes and thus reduces the overhead as shown in Figure 8.2. A shorter TTL
results in shorter TTLxR and invokes message forwarding quickly. In RSCRbase, messages
staying longer with regular nodes will not get replicated multiple times. Nonetheless,
messages with a lower TTL value are replicated as soon as contact opportunity occurs.
For a smaller TTL value, Epidemic attains a high delivery rate, and it decreases after an
increase in TTL.
• Case 2 – RSCRpred: RSCRpred can predict the link type and mean inter-contact time for
any forthcoming contact. We use Equation 8.15 to obtain the mean inter-contact time
and use it to determine the link type, as explained in Section 8.3.3.2 for any new con-
tact after time t. During the link determination process after t, initially, a link formed
between any two nodes is treated as sporadic, which increases the overhead. As Figure
8.3 shows, RSCRpred increases the delivery rate with less overhead as compared with
Epidemic, PRoPHET v2, and RSCRbase.
• Case 3 – RSCRpred: RSCRpred adjusts the α and T as a result of Case 2 to improve the
delivery rate further. With SLAW, RSCRpred alleviates the overhead with new values of α
and T. In Infocom ’05, the overhead by RSCRpred becomes less than PRoPHET v2’s and
Epidemic’s overheads under a higher TTL, as shown in Figure 8.3(b).

8.4 SUMMARY AND FUTURE DIRECTIONS

This chapter described the encounter- and contact-based routing methodology in MONs.
A brief review of many typical routing schemes was given, followed by a study of how
those schemes exploit the contact periodicity among mobile nodes to enhance the delivery
 Message Forwarding and Relay Selection  169

1 500
0.9 450 Epidemic
0.8 400 PRoPHETv2
0.7 350 RSCRbase
0.6 300

Overhead
Delivery

0.5 250
0.4 200
0.3 Epidemic 150
0.2 PRoPHETv2 100
0.1 RSCRbase 50
0 0
1 1.5 2 2.5 3 3.5 4 4.5 5 1 1.5 2 2.5 3 3.5 4 4.5 5
(a) Time to Live (Hours) (b) Time to Live (Hours)

Figure 8.2  Performance of RSCRbase (baseline: α = 0.3, T = 1 hour, t = 12 hours in SLAW).

1
0.9 28
0.8 24
0.7
20
Overhead

0.6
Delivery

0.5 16
0.4 Epidemic 12 Epidemic
0.3 PRoPHETv2 PRoPHETv2
8
0.2 RSCRpred RSCRbase
4
0.1 RSCRbase RSCRpred
0 0
1 1.5 2 2.5 3 3.5 4 4.5 5 1 2 3 4 5 6 7 8 9 10
(a) Time to Live (Hours) (b) Time to Live (Hours)

Figure 8.3  Performance of RSCRpred in comparison with RSCRbase and PRoPHETv2 (baseline: (a) α = 0.2, T =
2 hours, and t = 9 hours in SLAW, and (b) α = 4, T = 0.5 hour, and t = 12 hours in Infocom ’05).

Table 8.2  A summary of the topics of interest

Research Area Related Work Key Points

Social-based patterns [3, 4, 21] • Use of social relationship to build nodes’ betweenness
centrality with other nodes, social similarity, and tie strength
with the destination node.
• Nodes with higher centrality and similarity values are given
priority to receive a message.
Social-based contact [5, 6, 16, 20, 23] • Use of transient social contact patterns such as transient
patterns contact distribution, transient connectivity, and transient
community structure.
• Messages are forwarded using community structure and
transient contact patterns.
• Social-based contacts within community/communities to
calculate the centrality value and the use of frequency and
duration with respect to contacts.

(Continued)
170  Opportunistic Networks

Table 8.2  (Continued)

Research Area Related Work Key Points

Encounter-based [8, 9, 15, 19, 24] • Use of past encounter histories and inter-contact time to find
routing different patterns between two nodes.
• Use properties of movement behavior of different nodes to
select the next relay.
• Finding patterns using the correlation analysis to improve the
accuracy of prediction.
Periodicity-based [12, 14, 17, 22] • Use of some techniques like the auto-correlation function
routing (ACF) to find repetitive patterns and discrete Fourier
transform (DFT) to find distinct patterns.
• Capturing of skewed location, periodical reappearances of a
node.
• Use of social-based contact patterns from the temporal
perspective to predict future patterns.
• Use of inter-contact time to the destination node to select
the next relay node.

rate cost-effectively. This chapter explained the notion and finding of contact periodic-
ity and contact patterns between any node pair in a network. Contact patterns were then
used to determine the centrality metric of a node for its facility to communicate with other
nodes. Particularly, social and periodic features are used to calculate the centrality value of
each node or to find the strength with respect to a node pair. In addition, this chapter also
presented our recent study efforts on designing a new cost-effective contact-based routing
scheme, named RSCR, for message forwarding. The RSCR design can determine a link type
regular or sporadic between two nodes in a network. The proposed relay selection strategy
can use such link types together with TTL of a message to enhance the delivery rate with low
message overhead. Future directions in MONs attempt to integrate new trends, including
machine learning to predict future contacts, opportunistic offloading, device-to-device com-
munication, etc. With integration of new techniques and paradigms, the research community
in MONs will come up with novel solutions to resolve the uncertainty of contact predictabil-
ity and relay selection for cost-effective data delivery in MONs (Table 8.2).

REFERENCES

[1] Yue Cao and Zhili Sun. Routing in delay/disruption tolerant networks: A taxonomy, survey and
challenges. IEEE Communications Surveys Tutorials, 15(2):654–677, 2013.
[2] Amin Vahdat and David Becker. Epidemic routing for partially-connected ad hoc networks.
Technical report, Duke University, Durham, NC, 2000.
[3] E. M. Daly and M. Haahr. Social network analysis for information flow in disconnected delay-
tolerant MANETs. IEEE Transactions on Mobile Computing, 8(5):606–621, 2009.
[4] P. Hui, J. Crowcroft, and E. Yoneki. Bubble Rap: Social-based forwarding in delay-tolerant net-
works. IEEE Transactions on Mobile Computing, 10(11):1576–1589, 2011.
[5] K. Wei, D. Zeng, S. Guo, and K. Xu. On social delay-tolerant networking: Aggregation, tie detec-
tion, and routing. IEEE Transactions on Parallel and Distributed Systems, 25(6):1563–1573,
2014.
[6] F. Li, H. Jiang, H. Li, Y. Cheng, and Y. Wang. Sebar: Social-energy-based routing for mobile social
delay-tolerant networks. IEEE Transactions on Vehicular Technology, 66(8):7195–7206, 2017.
 Message Forwarding and Relay Selection  171

[7] Vijay Erramilli, Mark Crovella, Augustin Chaintreau, and Christophe Diot. Delegation forward-
ing. In Proceedings of the 9th ACM International Symposium on Mobile Ad Hoc Networking
and Computing, pp. 251–260, Hong Kong, China, May 2008.
[8] S. C. Nelson, M. Bakht, and R. Kravets. Encounter-based routing in DTNs. In Proceedings of
IEEE INFOCOM ’09, pp. 846–854, Rio de Janeiro, Brazil, April 2009.
[9] Sujata Pal and Sudip Misra. Contact-based routing in DTNs. In Proceedings of the 9th
International Conference on Ubiquitous Information Management and Communication, pp.
1–6, Bali, Indonesia, January 2015.
[10] Samo Grasic, Elwyn Davies, Anders Lindgren, and Avri Doria. The evolution of a DTN rout-
ing protocol – PRoPHETv2. In Proceedings of ACM CHANTS ’11, pp. 27–30, Las Vegas, NV,
September 2011.
[11] Thrasyvoulos Spyropoulos, Konstantinos Psounis, and Cauligi S. Raghavendra. Spray and
wait: An efficient routing scheme for intermittently connected mobile networks. In Proceedings
of the ACM SIG-COMM Workshop on Delay-Tolerant Networking (WDTN), pp. 252–259,
Philadelphia, PA, August 2005.
[12] Wei-Jen Hsu, T. Spyropoulos, K. Psounis, and A. Helmy. Modeling time-variant user mobility
in wireless mobile networks. In Proceedings of IEEE INFOCOM ’07, pp. 758–766, Barcelona,
Spain, May 2007.
[13] S. Hong Kyunghan Lee, S. J. Kim, I. Rhee, and S. Chong. Slaw: A new mobility model for human
walks. In Proceedings of IEEE INFOCOM ’09, pp. 855–863, Rio de Janeiro, Brazil, April 2009.
[14] Sungwook Moon and Ahmed Helmy. Understanding periodicity and regularity of nodal encoun-
ters in mobile networks: A spectral analysis. In Proceedings of IEEE GLOBECOM ’10, pp. 1–5,
Miami, FL, December 2010.
[15] Weitao Wang, Y. Bai, P. Feng, Y. Gu, S. Liu, W. Jiang, and J. Huang. DTN-KNCA: A high through-
put routing based on contact pattern detection in DTNs. In Proceedings of IEEE COMPSAC
’18, vol. 1, pp. 926931, Tokyo, Japan, July 2018.
[16] Xiaomei Zhang and G. Cao. Transient community detection and its application to data for-
warding in delay tolerant networks. IEEE/ACM Transactions on Networking, 25(5):2829–2843,
2017.
[17] Huan Zhou, Victor C. M. Leung, Chunsheng Zhu, Shouzhi Xu, and Jialu Fan. Predicting tem-
poral social contact patterns for data forwarding in opportunistic mobile networks. IEEE
Transactions on Vehicular Technology, 66(11):10372–10383, 2017.
[18] CRAWDAD Dataset. Mobile opportunistic network trace. Online; accessed October 25, 2020.
[19] Eyuphan Bulut, S. C. Geyik, and B. K. Szymanski. Efficient routing in delay tolerant networks
with correlated node mobility. In Proceedings of IEEE MASS ’10, pp. 79–88, San Francisco, CA,
November 2010.
[20] W. Gao, G. Cao, T. La Porta, and J. Han. On exploiting transient social contact patterns for data
forwarding in delay-tolerant networks. IEEE Transactions on Mobile Computing, 12(1):151–
165, 2013.
[21] Wei Gao, Qinghua Li, Bo Zhao, and Guohong Cao. Multicasting in delay tolerant networks: A
social network perspective. In Proceedings of ACM MobiHoc ’09, pp. 299–308, New Orleans,
LA, May 2009.
[22] Yu-Feng Hsu, C. Hu, and H. Hsiao. On exploiting temporal periodicity for message delivery in
mobile opportunistic networks. In Proceedings of IEEE COMPSAC ’18, vol. 1, pp. 809–810,
Tokyo, Japan, July 2018.
[23] X. Meng, G. Xu, T. Guo, Y. Yang, W. Shen, and K. Zhao. A novel routing method for social delay-
tolerant networks. Tsinghua Science and Technology, 24(1):44–51, 2019.
[24] M. Y. Mir and C. Hu. Exploiting mobile contact patterns for message forwarding in mobile
opportunistic networks. In IEEE Wireless Communications and Networking Conference
(WCNC), pp. 1–7, Seoul, Korea (South), May 2020.
[25] Ari Keränen, Jörg Ott, and Teemu Kärkkäinen. The one simulator for DTN protocol evaluation.
In Proceedings of Simutools ’09, pp. 55:1–55:10, Rome, Italy, March 2009.
[26] James Scott, Richard Gass, Jon Crowcroft, Pan Hui, Christophe Diot, and Augustin Chaintreau.
CRAWDAD dataset Cambridge/haggle (v. 2009-05-29), May 2009.
Chapter 9

Routing Techniques for Opportunistic

Network
Sonam Kumari
Raj Kumar Goel Institute of Technology, Ghaziabad, India
Itu Snigdh
Birla Institute of Technology, Mesra, India

CONTENTS

9.1 Introduction............................................................................................................ 173

9.2 Classification of Routing Algorithms...................................................................... 174
9.2.1 Context-Oblivious Approach........................................................................ 174
9.2.2 Context-Aware Approach............................................................................. 176
9.3 Factors Affecting Routing Protocol Performance.................................................... 178
9.4 QoS Constraints of MONs .....................................................................................182
9.5 ONE Simulator-Based Performance Evaluation of Routing Algorithms.................. 182
9.5.1 Case Study: Comparison of Average Latency................................................ 183
9.5.2 Analysis of the Routing Algorithms in View of Open Issues.......................... 183
9.6 Conclusion.............................................................................................................. 184
References....................................................................................................................... 184

9.1 INTRODUCTION

The mobile opportunistic network (MON) [5] is a challenged network as it is character-

ized by long and variable delays, high latency, frequent disconnections, long queuing times,
limited longevity, and limited resources. The connection is unstable between the nodes and
breaks as swiftly as it is discovered. The opportunistic network exploits human social behav-
ior such as mobility patterns and daily routine to accomplish routing of messages and sharing
data. With no fixed end-to-end path, the communication takes place only when opportunistic
contacts are established among mobile nodes. Therefore, it is tightly coupled with human
social behavior. Since the connection is highly dynamic, and since even the routes built here
are highly dynamic, traditional routing strategies in an opportunistic network cannot be used
efficiently.
With the abundance of mobile devices equipped with Wi-Fi, Bluetooth, sensors, and cam-
eras, several domains of research are being done on exploiting the advantages of the opportu-
nistic network. Moreover, modern vehicles are also employing such facilities and are installed
with communication interfaces and sensors. This situation is advantageous for communica-
tion as the number of available mobile devices in an environment, and hence the contact
opportunities, increase, which makes it useful in establishing an opportunistic network [6].
An MON carries a message created by a source node, with the help of intermediate nodes
that forward the message in such a way as to bring the message closer to the destination.
The communication model used in an MON is based on “store-carry-forwarding”. Due to
the variability in link existence and stability, nodes store the message on their buffer until

173
174  Opportunistic Networks

they establish contact with the next hop node and forward the message after discovering the
intermediate node [6].
The Internet we are using currently has certain assumptions on which it works, like end-
to-end connectivity, low round-trip times and two-way symmetry of link bandwidth, i.e.
uplink and downlink. Considering the case of an interplanetary communication taking place,
the propagation delay would be more if we use the usual Internet. There are no end-to-end
connectivity guarantees at a different instance of time – i.e. at a certain instance of time, two
nodes might be connected, but at another instance, they might not be connected, and a long
duration of time may elapse before those two nodes are connected again. The link bandwidth
for uplink and downlink might not be similar in such a network, and these networks are
highly error-prone.

9.2 CLASSIFICATION OF ROUTING ALGORITHMS

As we know that in an MON, a complete path between the sender and destination node
probably does not exist most of the time, we are not able to use traditional routing protocols
in such an environment. Therefore, designing an efficient routing protocol for opportunistic
networks is complex, as information about the network topology and the contact informa-
tion is insufficient. The existing routing strategies for the opportunistic network are broadly
categorized into two routing schemes: the context-oblivious approach, and the context-
aware approach. In a proactive protocol [7], every node in a network maintains a routing
table that is updated periodically and will broadcast the same to all other nodes available in
the network. The reactive routing protocol [7] is an “on-demand routing protocol”. In this
case, by flooding the Route REQuest (RREQ) packets on the network, every node dynami-
cally discovers its route to destination node.

9.2.1 Context-Oblivious Approach
The context-oblivious routing classification contains flooding-based routing protocols. These
routing protocols indiscriminately follow either flooding or controlled flooding strategies. In
the flooding-based strategy, a source node forwards packets to every neighboring node. Each
neighbor gets a packet and advances it to every one of its neighbors apart from the one from
which the packet was received. The destination node thus may receive many copies of that
packet through various nodes at an alternate time. The routing protocols of this category do
not exploit any form of context information of nodes, and the message is transmitted to the
destination by disseminating it as widely as possible. So, it is the simplest approach, as no
other prior information is calculated here for forwarding purposes. However, the flooding-
based strategy disperses an enormous number of packets in the network, which creates a
traffic issue, and it is also expensive as far as memory and energy utilization is concerned.
Epidemic routing and, Spray and Wait routing belong to this category.

• Direct transmission: Spyropoulos et al. [8] proposed a simple single-copy routing called
direct transmission routing. Direct transmission is a simple, single-copy routing proto-
col. In this routing protocol, the node which generates a message will hold that message
in its buffer until the time it reaches the destination node. In this scheme, data transfers
are minimized, but the delay in message delivery is high for this routing strategy and is
less robust.
• Epidemic routing: Vahdat and Becker [9] proposed epidemic routing for forwarding
data packets in an opportunistic network. It ensures that every node receives the packet
 Routing Techniques for Opportunistic Network  175

by a sufficient number of random exchanges of packets in the network. The dissemina-

tion process is bounded by providing hop count limits for each message.
• Spray and Wait: Spyropoulos et al. [10] proposed a technique called Spray and Wait,
which limits the flooding of packets, and thus the bandwidth requirement degrades in this
strategy. It basically consists of two phases, namely, the spray phase and the wait phase.
In the Spray phase, N copies of the message are created and are distributed to N number
of intermediate nodes randomly. Then, if the message is not delivered to the destination
node, N/2 numbers of copies are created and spread randomly. This process is repeated
until and unless one copy of the message is left out. Then in the Wait phase, direct trans-
mission is performed, and in this way, the packet is delivered to the destination node.
• Binary Spray and Wait based on Average Delivery Probability (ADPBSW): Xue et al.
[11] proposed another variation of the Spray and Wait routing protocol, Binary Spray
and Wait, which limits the flooding of packets by calculating average delivery probabil-
ity. Based on this, the next hop is selected for forwarding the packet towards the desti-
nation node. Whenever a node is encountered, it updates the delivery probability value.
In Spray and Wait, forwarding is done randomly, whereas in ADPBSW, forwarding is
based on delivery probability. If a message with destination node D is sent by node A
which possess N copies of this message and it encounters node B such that Pavg(b, d) >
N
Pavg(a, d), then node A will forward messages to node B. If Pavg(b, d) ≤ Pavg(a, d),
A will not forward messages to B. 2
• Network Coding protocol: Widmer and Boudec [12] proposed a network coding rout-
ing convention which is like flooding-based routing or probabilistic routing, except it
utilizes network coding to restrict message flooding. The overhead of probabilistic rout-
ing protocol is reduced in the Network Coding protocol. Here, each message is linked
with an encoding vector, and the message vector is transmitted along with the encoding
vector. A decoding matrix is used to store the message and encoding vector, as well as
the original message vector in case it is a source node. So, in this protocol, the message
is transformed by the nodes in another format before transmission. The performance
of this protocol is higher than the probabilistic-based protocol and is suitable for the
extreme environment such as sparse mobile networks.
• Erasure-coding based routing: Wang et al. [13] proposed this protocol where forward-
ing of data is based on erasure codes. The longer delays are reduced, as it allows to use
a larger number of intermediate nodes while maintaining the overhead. The message
is converted into code blocks with the end goal that any adequately enormous subset
of the created code blocks can be utilized to reproduce the original message. It takes
as information a message of size M and a replication factor r. The calculation pro-
duces M × r/b equivalent measured code blocks of size b, with the end goal that any
M
(1 + €). code blocks can be utilized to recreate the message. Here, € is a constant and
2
differs relying upon the specific algorithm utilized; for example, Reed-Solomon codes
or Tornado codes. The determination of the algorithm includes compromises between
coding/decoding effectiveness and the minimal number of code blocks required to
remake a message.
• Hybrid erasure-coding based routing (H-EC): Chen et al. [14] proposed the H-EC
protocol which is intended to completely consolidate the robustness of the erasure
coding-based protocol while saving the performance of replication techniques. In this
methodology, two duplicates of eradication-coded blocks are forwarded to intermedi-
ate nodes. The first duplicate of coded blocks is sent by the Erasure coding way such
that each relay node gets the coded block, and the second duplicate of erasure coded
blocks is communicated by the Aggressive Erasure Coding protocol.
176  Opportunistic Networks

9.2.2 Context-Aware Approach
Context is any relevant information that describes the environment of any objects (i.e. a
person, groups, and physical and computational devices) that interact. In the knowledge-
based approach [15–17], based on certain information related to the environment, a node is
selected as an intermediate node such that it increases the delivery probability and makes a
faster delivery possible. The information may be related to either network topological order,
network traffic, or node location.

• Randomized routing: Spyropoulos et al. [8] proposed this single copy routing protocol
where it uses intermediate nodes to deliver the message to the destination node. Here,
a node transmits the message to the intermediate node it encounters with probability
p > 0 only if its probability of delivering the message to the destination node is greater
than the present node holding the packet. The time of contact information is used to
calculate the delivery probability.
• Utility-based routing: Spyropoulos et al. [8] proposed this single copy routing protocol
where the position data regarding each node is updated in the last encounter timer and
is diffused in the network though the mobility process of other nodes. This position
data is not absolute, but rather relative to the position of another node. If a node is seen
at some time instant having a low timer value for another node, then this other node
is expected to be somewhere nearby. This protocol is the variation of the randomized
routing protocol, as the speed, mobility pattern, and last encountered time information
are used to calculate the delivery probability.
• Seek and Focus routing protocol: Spyropoulos et al. [8] proposed the Seek and Focus
protocol. A utility-based protocol has the disadvantage of a slow-starting initial stage,
which can be more prevalent in larger networks. Particularly in a huge network, where
anticipated separation between a source and a destination node is greater, it will take the
source node quite a while to find a higher-utility next hop at the start. The Seek and Focus
protocol consists of two stages. In the Seek phase, if the utility around the hub is low,
perform randomized forwarding with boundary p to rapidly look for close-by hubs; and
in the Focus phase, when a high utility hub is found, change to a utility-based protocol.
• Prioritized Epidemic routing: Ramanathan et al. [18] proposed PRioritized EPidemic
routing (PREP), where the priority of the packet is dependent on cost to destination,
source, and expiry time. The value of cost is dependent on the per-link “average avail-
ability” information that is spread in the epidemic manner. Here every node executes
a neighbor discovery algorithm to create and maintain a set of bidirectional links with
neighboring nodes. Each message is allocated with drop priority pd and transmit prior-
ity pt. The packet whose transmit priority is high will be forwarded first to the node
who comes in its contact. When the buffer is full than the high drop priority packet will
be deleted first than the packet with low drop priority. PREP uses expiry time informa-
tion and topology awareness to take a decision of deleting or holding the packet in a
scenario where resources (buffer, bandwidth) are fully consumed.
• Spray and Focus routing protocol: Spyropoulos et al. [15] proposed the Spray and
Focus protocol in which packet replication is reduced and distributed among a few
relay nodes. Further, the relay node will forward the packet using a single copy utility-
based scheme and carry the packet until time to live expires. Since many relay nodes
are in parallel looking for the destination node, it becomes an efficient protocol in the
sparse connectivity environment. In the spray phase, the number of copies of the mes-
sage generated “L” is such that the average delay is α times the optimal one where M
number of nodes are present, and it is calculated using the following equation:
 Routing Techniques for Opportunistic Network  177

 2 2  2  2M  1  M
H 3
M 
 1.2 L3   H M 
6
 L   a 
M  M  1
 L 
 M 1
(9.1)
  
In the focus phase, the message can be forwarded to a different intermediate node
based on forwarding criteria which is based on a set of timers that record the time since
two nodes last encountered each other. So, here, every node keeps up a vector with
IDs of all messages that it has stored, and at whatever point two nodes encounter one
another, they exchange their vectors and check which messages they have in common
and their time to live (TTL). If the TTL expires, the message is discarded.
• Motion vector (MoVe): LeBrun et al. [19] introduced a scheme using the motion vector
(MoVe) of mobile nodes to predict their future location. In this strategy information
of nodes, i.e. relative velocity, is calculated, and using this information, a node’s future
location is predicted and this gives its neighboring nodes information such that they
bring the message closer to the destination. The data success rate of MoVe as a function
of the number of nodes is less than Epidemic routing, and an end-to-end delay is longer
than epidemic routing.
• Bubble Rap routing protocol: Hui and Crowcroft [20] proposed the Bubble Rap proto-
col, which is based on the concept of community such that the effectiveness of forward-
ing strategy is improved. It follows two significant observations. First is the popularity
of the node in the community, so the packet is forwarded to the most popular node.
Second is the community to which the relay node belongs; if the relay node is of the
same community as that of destination node, there is a greater chance that the interac-
tion between the nodes will take place. There are a few assumptions of the protocol:
Each node must belong to at least one community, and a community can consist of a
single node. Every node has two types of rankings: global ranking and local ranking.
Global ranking represents the popularity of the node across the network, whereas local
ranking indicates only the popularity of the node within the community to which it
belongs. The source node constructs the hierarchical ranking tree based on global rank-
ing until it reaches a node which belongs to the destination node community. After that,
it uses a local ranking tree to deliver the message to the destination node.
• Context-Aware Routing (CAR): Musolesi et al. [21] presented Context-Aware Routing
(CAR). In CAR, the message is transmitted to a node that has the highest delivery prob-
ability. The delivery probability is calculated and predicted using some context informa-
tion with the help of the Kalman filter. Delivery probabilities are exchanged periodically
so that the best carriers are computed based on the nodes’ context information.
• MaxProp: Burgess et al. [22] proposed a protocol called MaxProp for effective routing
of messages. In this routing strategy, a node schedules the packets for transmission to
its neighboring nodes based on path likelihoods which are predicted using the history
of contacts. This scheme also uses few mechanisms, such as providing an acknowledg-
ment, a head-start for new packets, and lists of previous intermediaries used in order to
enhance the prediction
• Shortest expected path routing (SEPR): Kun et al. [23] proposed a shortest expected
path routing (SEPR). SEPR estimates the probability of link forwarding using the his-
torical information. Effective path length (EPL) is estimated in this scheme, and when-
ever two nodes meet, they exchange this information with each other and update its
EPL value in the link probability table. The smaller the EPL value, the higher the chance
of successful message delivery. Nodes recalculate their probability to meet other nodes
when they get connected. The probability is calculated basically using the time those
two nodes stay connected divided by the sampling time window. The Dijkstra shortest
path algorithm is used to get the expected path length for each node in the network.
178  Opportunistic Networks

• Estimate/prediction routing: Lindgren et al. [24] proposed a probabilistic routing pro-

tocol called PRoPHET (Probabilistic Routing Protocol using History of Encounters and
Transitivity). PRoPHET is based on the estimation of delivery probability. This rout-
ing strategy works similar to the Epidemic routing. In real life, humans mostly move
predictably according to their repeating behavioral patterns. For example, if a user has
visited a spot ordinarily previously, it is likely that the user will visit that place once
more. In view of these perceptions, the convention utilizes the recurrence of contacts
between the nodes as setting data to improve the routing performance. Whenever nodes
meet, they exchange tables containing the delivery probability value. It is also observed
that if two nodes meet each other frequently, they have a high delivery probability to
each other. Hence, the delivery probability ages with time.
• History-based prediction routing (HBPR): Dhurandher et al. [25] proposed HBPR,
which uses the behavioral information of the nodes to find the next node for forward-
ing the packet. A few assumptions are taken into consideration:
- The movement of the nodes are based on the Human Mobility pattern
- The nodes are cooperative
- The simulated area is termed as cells, and they can be locations that nodes visit dur-
ing simulation process.
The protocol is divided into three phases:
• Initializing the Home Locations
• Message generation and Home Location update
• Next hop selection.
Each node has a certain region that it would visit frequently, some regions that it would
visit but not often, and so forth. Based on this, one could predict a node’s future loca-
tion quite effortlessly while using its past history. Every node floods their past location
in the network and keeps updating their history after performing a re-scan process.
• kROp: Deepak et al. [26] proposed the k-Means clustering based routing protocol.
kROp utilizes unsupervised machine learning in the form of an optimized k-Means
clustering algorithm to train on these features and make next-hop-selection decisions.
This protocol is based on Feature identification, a Clustering model, and an Evaluation
function. In the Feature identification phase, the nodes’ information, such as encounters
of the node with the destination node, distance of the node from the destination node,
available buffer space of the node, and number of messages successfully delivered by
that node until that point of time, are used for next-hop prediction. An evaluation func-
tion is defined that helps to select the best cluster among the cluster centers generated.

Table 9.1 compares the context oblivious routing protocols, and Table 9.2 compares context-
aware protocols on the basis of buffer management, estimation of forwarding probability,
reliability, energy efficiency, and types of routing whether reactive or proactive.

9.3 FACTORS AFFECTING ROUTING PROTOCOL PERFORMANCE

The key problem in the opportunistic network is to choose to the appropriate relay nodes
to transfer the message such that the message is delivered with minimum delay latency and
maximum probability of successful delivery of the message to the destination node. The
parameters that affect the overall performance of routing protocol are:

• Social reference
• Delivery probability
• Node speed
• Node remaining energy
 Routing Techniques for Opportunistic Network  179

Table 9.1  Comparison between context oblivious routing protocols.

Protocols Buffer Estimation of Reliability Energy Reactive or
Management Forwarding Efficiency Proactive
Probability

Direct Infinite No More reliable More Reactive

transmission (high traffic) bandwidth is
required
Epidemic Infinite No Less reliable (high More Reactive
traffic serves bandwidth is
high contention) required
Spray and Wait Infinite No Less reliable Less Reactive
(similar to bandwidth
Epidemic) than
Epidemic
Binary Spray and Infinite No; spreads Less than Less Reactive
Wait based on copies of epidemic bandwidth
Average Delivery message in compared
Probability controlled way with
Epidemic
Network Coding Infinite No, merges Less than Less Reactive
protocol several epidemic bandwidth
messages into compared
one to reduce with
number of Epidemic
transmissions
Erasure-coding Infinite No, converts Less than Less Reactive
based routing message into epidemic bandwidth
several code compared
blocks with
Epidemic
Hybrid erasure- Infinite No, combination Less than Less Reactive
coding based of EC and A-EC epidemic bandwidth
routing compared
with
Epidemic

Table 9.2  Comparison between context-aware routing protocols.

Protocols Buffer Estimation of Reliability Energy Efficiency Reactive or
Management Forwarding Proactive
Probability

Randomized No Last encountered Less reliable Less bandwidth Reactive

routing time
Utility-based No Last encountered Less reliable Less bandwidth Reactive
routing time with speed
and mobility
pattern
Seek and No Last encountered Less reliable than Less bandwidth Reactive
Focus routing time with speed epidemic than epidemic
protocol and mobility
pattern

(Continued)
180  Opportunistic Networks

Table 9.2­  (Continued)

Protocols Buffer Estimation of Reliability Energy Efficiency Reactive or

Management Forwarding Proactive
Probability

Prioritized Yes Use expiry time Less reliable than Less bandwidth Reactive
Epidemic information epidemic than epidemic
routing and topology
awareness
Spray and Infinite Last encountered Reduces delay Less bandwidth Reactive
Focus routing time of Spray and than epidemic
protocol Wait Routing up
to 20 times in
some scenarios
MoVe Infinite Yes, using the Less than Less bandwidth Reactive
motion vector epidemic with
substantially
less buffer space
compared with
epidemic
Bubble Rap Yes, infinite Social Better Less bandwidth Reactive
routing community forwarding than epidemic
protocol information efficiency
compared
to context-
oblivious
routing and
PROPHET
routing.
CAR Infinite Yes, using Kalman More reliable Less bandwidth Reactive
filter than epidemic if than an
the buffer size is epidemic
small
MaxProb Infinite Yes, estimating More reliable Less bandwidth Reactive
the delivery than CAR than CAR
likelihood (higher delivery
rate)
SEPR Yes, removes Yes More reliable Less bandwidth Reactive
those than epidemic than epidemic
packets with (almost 35 (almost 50
smaller EPL percent) percent)
PRoPHET Infinite Yes, using More reliable Less bandwidth Reactive
delivery than epidemic than an
probability epidemic
vector
History-based Infinite Yes, using nodes More reliable Less bandwidth Reactive
prediction mobility pattern than epidemic than epidemic
routing
kROp Infinite Yes, utilizes More reliable Less bandwidth Reactive
unsupervised than HBPR than HBPR
machine
learning
 Routing Techniques for Opportunistic Network  181

Social reference: The MON mimics human social behavior, and this is estimated on the
value of betweenness centrality [27]. Betweenness centrality of a node is the ratio
between the numbers of the shortest path that passes through that node to the total
number of the shortest path:

st   
Betweenness    s    t (9.2)
st
where σst is the total number of shortest paths from node s to node t and σst(ѵ) is the num-
ber of those paths that pass through ѵ.
Delivery probability: The delivery ratio (DR) is directly proportional on the two factors,
contact duration (CT) and inter-contact time (ICT) [3]. Contact duration is the time
period for which the devices are in contact with each other. Intercontact time is the time
interval after which the two devices will come in contact with each other.

Contact Duration
DR α (9.3)
Inter Contact Time

Node speed: Normally, the user carries the mobile device with them, so the node speed
parameter is dependent on transportation mode of the user. In general, it has been
observed that the speed of a walking person varies from 0.5 to 1.5 meters per second,
the speed of person traveling by a vehicle is assumed to vary from 8.3 to 30 meters per
second and of a cyclist is assumed to vary from 4.2 to 11 meters per second.
Node remaining energy: Node remaining energy (NRE) inversely depends on the number
of packets it transmits and receives and the inter-probe time [28]. As the NRE value
decreases, the likeliness of the node suffering downtime increases and the probability of
node being used as the next hop decreases.
1
NRE α (9.4)
Noof packets received and transmitted

E  Send  k, d    Eelec  k  amp  k  dij

2
(9.5)

E  Recv  k    Eelec  k (9.6)

where Eelec is the coefficient of radio frequency, εamp is the amplification coefficient of
sending the device, dij is the data transmission distance of the node, and k is the number
of bits being transferred.
Inter probe is the time interval after which the node will start the neighbor discovery
process again:

1
Inter probe time     (9.7)
fs

2B
Where fs =
R

where B is the speed at which the node travels in meter per second. R is the transmission
radius of the node in the meter. The variable α lies within the range 0 < α ≤ 1 and it
defines the degree of closeness of nodes to be classified as its neighbors. Generally, the
182  Opportunistic Networks

transmission range of Bluetooth is 30 meters and that of a Wi-Fi device is 50 meters.

Therefore, let’s say a person is carrying a Bluetooth-enabled mobile phone in his pocket
and traveling in a vehicle at a speed of 25 meters per second. The inter probe time cal-
culation is:

2B
fs =
R

After substituting, R = 30 meters, B = 25 meters/second, and fs = 1.67 per second. Therefore,

Inter probe time (τ) = 1/(αfs) = 0.59 second (assuming α = 1).

9.4 QoS CONSTRAINTS OF MONs

Though the opportunistic network seems to be the available solution to the incessant mobil-
ity-based communication, deploying an opportunistic network in the real environment poses
several challenges. The two major challenges of an opportunistic network are:

• Opportunistic contact
• Storage constraint
Contact: Since in such environments, nodes are mobile, the contact between nodes
becomes unpredictable. The contact duration and inter-contact time are important
factors through which we can determine the capacity of the opportunistic network.
Contact duration is the time period for which the two devices are in contact with
each other and the inter-contact time is the time interval after which the contact
between the same pair of devices takes place.
Storage constraint: In order to avoid packet drop in an opportunistic network, the
packets must be stored in the intermediate nodes buffer space for the time duration
the packet is delivered to the appropriate destination node. As the number of mes-
sages in a network increases, the buffer requirement increases. We can say that the
buffer requirement is directly proportional to a number of messages created in the
network; for example, the epidemic routing protocol, which, based on the flooding
approach, needs unlimited relay buffers. Routing can be improved if we are able to
gather more information such as network expected topology and contact informa-
tion. Unfortunately, such data are not easily available.

9.5 ONE SIMULATOR-BASED PERFORMANCE EVALUATION OF

ROUTING ALGORITHMS

There exist several simulation tools for the delay tolerant network, such as dtnsim, dtm-
sim2, ns2, and OMNeT for MANET, through which various approaches can be tested and
evaluated by performing simulation under a certain scenario. Similarly, for the opportu-
nistic network, a simulation environment called the Opportunistic Network Environment
(ONE) simulator is available. The ONE simulator not only focuses on routing but also
combines various mobility model and visualization in one package that is easily extensible
and provides a rich set of reporting and analyzing modules. The ONE simulator incorpo-
rates various types of movement models proposed for the opportunistic network, such as
map-based movement, Random Waypoint, Random Walk, and shortest path map-based
movement.
 Routing Techniques for Opportunistic Network  183

9.5.1 Case Study: Comparison of Average Latency

Latency is the time delay of a packet to reach from the source to the desired destination
node. Simulation could be performed for several existing routing protocols. In a case study,
Epidemic routing, Spray and Wait routing, and PRoPHET routing simulations were per-
formed. The observation was based on different numbers of nodes, and the average latency
of these routing protocols has been compared and plotted in the graph shown in Figure 9.1.
From the graph, it has been observed that the average latency of PRoPHET routing is less
than that of the other two routing protocols.

9.5.2 Analysis of the Routing Algorithms in View of Open Issues

Opportunistic networks present a lot of challenges owing to their mobility and disconnected
behavior. As in the previous section, simulation was performed for comparing the average
latency of different protocols, in a similar way we can also evaluate other factors (i.e. average
buffer time, overhead ratio) based on which of several routing protocols can be compared.
Buffer space is the memory space required to store the packet into nodes memory. Packets
are stored in an intermediate nodes memory until it transmits the packet to another node.
Packet buffering is important for faster delivery and to reduce the packet loss problem. The
buffer management algorithm is used in order to schedule the packet and to take a decision
whether to accept or reject the packet. For the average buffer, time simulation was performed.
In Figure 9.2 a graph of average buffer time versus a number of nodes for the PRoPHET,
Epidemic, and Spray and Wait routing protocols has been plotted. It is observed that the aver-
age buffer time of PRoPHET is less than that of Epidemic and Spray and Wait as the number
of nodes increases.

Figure 9.1  Comparison graph of average latency of routing protocols.

184  Opportunistic Networks

Figure 9.2  Comparison graph of average buffer time of routing protocols.

9.6 CONCLUSION

We have outlined the frequently used routing protocols for opportunistic networks. We now
understand that the success of any network is the capability to efficiently and reliably com-
municate data. The requirements of an MON are different from those of the regular wireless
networks wherein we can leverage the mobility of nodes to our advantage. Also, in wireless
networks, the increase in node density cause interference and packet drops. On the contrary,
the performance of MONs improves with the increase in the number of nodes. So, with the
increase in the number of communicating nodes, the chances of forwarding increase, and
hence the delay and the required buffer time decreases.

REFERENCES

[1] Lindgren, Anders, and Avri Doria. “Experiences from deploying a real-life DTN system.” In
IEEE Consumer Communications and Networking Conference: 11/01/2007-13/01/2007, pp.
217–221. IEEE Communications Society, 2007.
[2] Huang, Chung-Ming, Kun-chan Lan, and Chang-Zhou Tsai. “A survey of opportunis-
tic networks.” In 22nd International Conference on Advanced Information Networking and
Applications-Workshops (AINA Workshops 2008), pp. 1672–1677. IEEE, 2008.
[3] Pelusi, Luciana, Andrea Passarella, and Marco Conti. “Opportunistic networking: Data for-
warding in disconnected mobile ad hoc networks.” IEEE communications Magazine 44, no. 11
(2006): 134–141.
[4] Santos, Rodrigo M., Javier Orozco, Sergio F. Ochoa, Roc Meseguer, and Daniel Mosse. “Providing
real-time message delivery on opportunistic networks.” IEEE Access 6 (2018): 40696–40712.
[5] Conti, Marco, Silvia Giordano, Martin May, and Andrea Passarella. “From opportunistic
networks to opportunistic computing.” IEEE Communications Magazine 48, no. 9 (2010):
126–139.
[6] Kumari, Sonam, and Itu Snigdh, “Analyzing impact of factors on routing decisions in opportu-
nistic mobile networks,” Conference Publications, ICRTCST – 2018.
 Routing Techniques for Opportunistic Network  185

[7] Kumar, Manish, Itika Gupta, Sudarshan Tiwari, and Rajeev Tripathi. “A comparative study of
reactive routing protocols for industrial wireless sensor networks.” In International Conference
on Heterogeneous Networking for Quality, Reliability, Security and Robustness, pp. 248–260.
Springer, Berlin, Heidelberg, 2013.
[8] Spyropoulos, Thrasyvoulos, Konstantinos Psounis, and Cauligi S. Raghavendra. “Single-
copy routing in intermittently connected mobile networks.” In 2004 First Annual IEEE
Communications Society Conference on Sensor and Ad Hoc Communications and Networks,
2004. IEEE SECON 2004, pp. 235–244. IEEE, 2004.
[9] Vahdat, Amin, and David Becker. “Epidemic routing for partially connected ad hoc networks.”
(2000): 2019.
[10] Spyropoulos, Thrasyvoulos, Konstantinos Psounis, and Cauligi S. Raghavendra. “Spray and
wait: An efficient routing scheme for intermittently connected mobile networks.” In Proceedings
of the 2005 ACM SIGCOMM Workshop on Delay-Tolerant Networking, pp. 252–259. 2005.
[11] Xue, Jingfeng, Xiumei Fan, Yuanda Cao, Ji Fang, and Jiansheng Li. “Spray and wait rout-
ing based on average delivery probability in delay tolerant network.” In 2009 International
Conference on Networks Security, Wireless Communications and Trusted Computing, vol. 2, pp.
500–502. IEEE, 2009.
[12] Widmer, Jörg, and Jean-Yves Le Boudec. “Network coding for efficient communication in
extreme networks.” In Proceedings of the 2005 ACM SIGCOMM Workshop on Delay-Tolerant
Networking, pp. 284–291. 2005.
[13] Wang, Yong, Sushant Jain, Margaret Martonosi, and Kevin Fall. “Erasure-coding based routing
for opportunistic networks.” In Proceedings of the 2005 ACM SIGCOMM Workshop on Delay-
Tolerant Networking, pp. 229–236. 2005.
[14] Chen, Ling-Jyh, Chen-Hung Yu, Tony Sun, Yung-Chih Chen, and Hao-hua Chu. “A hybrid rout-
ing approach for opportunistic networks.” In Proceedings of the 2006 SIGCOMM Workshop on
Challenged Networks, pp. 213–220. 2006.
[15] Spyropoulos, Thrasyvoulos, Konstantinos Psounis, and Cauligi S. Raghavendra. “Spray and
focus: Efficient mobility-assisted routing for heterogeneous and correlated mobility.” In Fifth
Annual IEEE International Conference on Pervasive Computing and Communications
Workshops (PerComW'07), pp. 79–85. IEEE, 2007.
[16] Verma, Anshul, K. K. Pattanaik, Aniket Ingavale, “Context-Based Routing Protocols for
OppNets”, Routing in Opportunistic Networks, New York, Springer, chapter 3, pp. 69–97,
2013. doi:10.1007/978-1-4614-3514-3_3
[17] Verma, Anshu, K. K. Pattanaik, “Routing Protocols in Opportunistic Networks.” Opportunistic
Networking: Vehicular, D2D and Cognitive Radio Networks, CRC Press (Taylor & Francis
Group), chapter 5, pp. 125–166, 2017. doi:10.1201/9781315200804.
[18] Ramanathan, Ram, Richard Hansen, Prithwish Basu, Regina Rosales-Hain, and Rajesh
Krishnan. “Prioritized epidemic routing for opportunistic networks.” In Proceedings of the 1st
International MobiSys Workshop on Mobile Opportunistic Networking, pp. 62–66. 2007.
[19] LeBrun, Jason, Chen-Nee Chuah, Dipak Ghosal, and Michael Zhang. “Knowledge-based
opportunistic forwarding in vehicular wireless ad hoc networks.” In 2005 IEEE 61st Vehicular
Technology Conference, vol. 4, pp. 2289–2293. IEEE, 2005.
[20] Hui, Pan, and Jon Crowcroft. Bubble Rap: Forwarding in small world DTNs in ever decreasing
circles. No. UCAM-CL-TR-684. University of Cambridge, Computer Laboratory, 2007.
[21] Musolesi, Mirco, and Cecilia Mascolo. “CAR: Context-aware adaptive routing for delay-toler-
ant mobile networks.” IEEE Transactions on Mobile Computing 8, no. 2 (2008): 246–260.
[22] Burgess, John, Brian Gallagher, David D. Jensen, and Brian Neil Levine. “MaxProp: Routing for
vehicle-based disruption-tolerant networks.” In Infocom, vol. 6, pp. 1–11. 2006.
[23] Tan, Kun, Qian Zhang, and Wenwu Zhu. “Shortest path routing in partially connected ad hoc
networks.” In GLOBECOM'03. IEEE Global Telecommunications Conference (IEEE Cat. No.
03CH37489), vol. 2, pp. 1038–1042. IEEE, 2003.
[24] Lindgren, Anders, Avri Doria, and Olov Schelén. “Probabilistic routing in intermittently con-
nected networks.” ACM SIGMOBILE Mobile Computing and Communications Review 7, no.
3 (2003): 19–20.
186  Opportunistic Networks

[25] Dhurandher, Sanjay K., Deepak Kumar Sharma, Isaac Woungang, and Shruti Bhati. “HBPR: his-
tory based prediction for routing in infrastructure-less opportunistic networks.” In 2013 IEEE
27th International Conference on Advanced Information Networking and Applications (AINA),
pp. 931–936. IEEE, 2013.
[26] Sharma, Deepak Kumar, Sanjay Kumar Dhurandher, Divyansh Agarwal, and Kunal Arora.
“kROp: k-Means clustering based routing protocol for opportunistic networks.” Journal of
Ambient Intelligence and Humanized Computing 10, no. 4 (2019): 1289–1306.
[27] Liu, Tong, Yanmin Zhu, Ruobing Jiang, and Bo Li. “A sociality-aware approach to computing
backbone in mobile opportunistic networks.” Ad Hoc Networks 24 (2015): 46–56.
[28] Orlinski, Matthew, and Nick Filer. “Neighbour discovery in opportunistic networks.” Ad Hoc
Networks 25 (2015): 383–392.
Chapter 10

Blockchain Leveraged Node Incentivization

in Cooperation-Based Delay Tolerant
Networks
Souvik Basu
Heritage Institute of Technology, Kolkata, India

Soumyadip Chowdhury
University of Engineering and Management, Kolkata, India

Siuli Roy
Heritage Institute of Technology, Kolkata, India

CONTENTS

10.1 Introduction............................................................................................................. 188

10.2 Literature Review.................................................................................................... 189
10.3 Rudimentary Elements of Blockchain...................................................................... 190
10.3.1 Blocks.......................................................................................................... 190
10.3.2 Transaction .................................................................................................191
10.3.3 Mining......................................................................................................... 191
10.3.4 Ethereum and Smart Contracts.................................................................... 191
10.4 Integration of Blockchain with DTNs...................................................................... 192
10.5 System Model.......................................................................................................... 193
10.5.1 Network Architecture.................................................................................. 193
10.5.2 Security Model ............................................................................................194
10.6 BlockCent: Blockchain-Based Node Incentivizing Scheme........................................ 194
10.6.1 Setup Phase.................................................................................................. 195
10.6.2 Incentivizing Process ...................................................................................195
10.6.2.1 Transaction Creation .................................................................. 195
10.6.2.2 Message Transmission................................................................. 196
10.6.2.3 Incentive Redemption................................................................. 196
10.6.3 Reward Model............................................................................................. 197
10.7 Security Analysis ..................................................................................................... 198
10.7.1 Shelter Node Refuses to Pay Back............................................................... 198
10.7.2 Forwarder Node Refuses to Forward Message ............................................198
10.7.3 Forwarder Node Refuses to Send ACK........................................................ 198
10.7.4 Observer Node Poses as Forwarder Node.................................................... 198
10.8 Experimental Results............................................................................................... 199
10.8.1 Simulation Environment .............................................................................199
10.8.2 Simulation Setup .........................................................................................199
10.8.2.1 ONE Setup ................................................................................. 199
10.8.2.2 Ethereum Setup........................................................................... 200
10.8.3 Simulation Metrics....................................................................................... 200
10.8.3.1 Design Metrics............................................................................ 201
10.8.3.2 Network Performance Metrics.................................................... 201

187
188  Opportunistic Networks

10.8.3.3 Blockchain Metrics..................................................................... 201

10.8.4 Results and Discussion.................................................................................202
10.8.4.1 Achieving Major Design Goals................................................... 202
10.8.4.2 Evaluation of Network Performance ........................................... 203
10.8.4.3 Overheads Introduced for Blockchain Activities.......................... 204
10.9 Conclusion............................................................................................................... 204
References........................................................................................................................ 205

10.1 INTRODUCTION

A DTN, which is essentially an opportunistic peer-to-peer network, involves the transmis-

sion of messages from a sender to the receiver in multiple hops through forwarder nodes.
Thus, the success of such a network is highly dependent on the cooperation of these for-
warder nodes, and selfish activities contribute to low delivery ratio and long latency of
delivery [1–6]. Intentions behind such selfish behavior may be either non-malicious (con-
cerns about energy and bandwidth consumption, etc.) or malicious (attacks like denial of
service, collusion, impersonation, etc.) [5]. One solution to this problem is to reward for-
warder nodes with real money or credit to stimulate them to engage in forwarding activi-
ties. The characteristics of an opportunistic peer-to-peer network, however, bring challenges
to the creation of an incentive mechanism. On one hand, how can a sender be convinced
that the message, containing shelter-needs, is actually delivered to the receiver? If a sender
rewards a selfish node beforehand, the node will not faithfully store-carry-forward the mes-
sage to the receiver and will indulge in a “dine and dash” behavior [7]. On the other hand,
a sender does not know who the forwarder nodes will be, and how many such nodes will
be involved in transmitting a message; thus, it is hard to know who will be rewarded and
how much reward is to be paid [8]. Finally, forwarder nodes may have different constraints
on consumed resources and different expectations about the incentive. A node will not
diligently store-carry-forward the message to the receiver and engage in “dine and dash”
behavior if a sender rewards a greedy node beforehand. On the other hand, a sender does
not know who the forwarder nodes will be and how many such nodes will be involved in
the transmission of a message; it is therefore difficult to know who will be rewarded and
how much reward will be paid [8]. Finally, forwarder nodes can have distinct resource con-
straints and different expectations.
Several credit-based approaches [9–14] to developing effective incentive structures for
DTNs have been published, where rewards are given to the cooperative forwarder nodes
through paying credit or virtual money. Most credit-based reward schemes, however, either
rely on or do not have an explicit digital currency framework that is proven to be secure [8].
Blockchain [15–18] is a distributed digital public transaction leader of cryptocurrencies such
as Bitcoin or Ethereum [17, 18] which, in lieu of a central trusted authority, can verifiably
and permanently record transactions between two parties. The most realistic digital currency,
Ethereum, makes it possible for P2P networks to formulate practical credit-based reward
schemes. However, the usage of blockchain is restricted by the user’s access to an end-to-end
Internet connection. This limitation restricts the use cases to access blockchain and prevents
its adoption in intermittently connected network environments like DTNs. In fact, reliance
on the Internet is becoming a critical concern for blockchain developers, since network inter-
ruptions can cause serious harm to the blockchain ecosystem. Thus, it becomes imperative to
look for new avenues through which blockchain can be used even in intermittently connected
network environments. To the best of our knowledge, this domain is much less investigated.
This chapter first proposes a mechanism that enables usage of blockchain technology in an
intermittently connected network such as DTN based on the Ethereum platform using smart
 Blockchain Leveraged Node Incentivization  189

contracts in Solidity. The mechanism develops an alternative way of broadcasting and autho-
rizing blockchain transactions without having to rely on Internet connectivity. It exploits
the store-carry-forward feature of DTN nodes in conjunction with stationary relay nodes
(e.g., smartphone, laptop), called DropBoxes to transmit and authorize transactions, hence
enabling the usage of blockchain in DTN. The mechanism successfully integrates blockchain
with DTN to exploit the immutability and accessibility of blockchain in an intermittently
connected network. The chapter next proposes BlockCent, a blockchain-based node incen-
tivizing scheme for a DTN-leveraged, smartphone-based, peer-to-peer network that runs on
the blockchain-DTN integrated environment. In this scheme, forwarder nodes who help in
successful transmission of messages are given rewards, in terms of Ethereum. Exploiting the
Ethereum cryptocurrency enables us to design the incentive scheme at low cost, alleviat-
ing concerns about the reliability of virtual coin rewards in DTNs. Unlike the conventional
blockchain system, the parties that make transactions, i.e. the forwarder nodes, do not need
direct access to the blockchain network in our system. The scheme uses a novel reward model
to bring rationality to the incentivizing process.
The proposed incentivizing scheme is applied to the disaster management use case.
When a disaster strikes, collection of the victims’ needs at different relief shelters and their
transmission to the emergency operation center (EOC) is crucial. These need information
are used by different stake holders (e.g. relief agencies, government agencies, NGOs, and
disaster managers) appropriately coordinate and manage relief activities [1–5]. Since large
portions of communication facilities get incapacitated as a result of the disaster, it gets
hard to gather and communicate the needs from various remote and inaccessible shel-
ters. The networking research group has strongly suggested the use of delay tolerant net-
works (DTNs) to set up a post-disaster communication infrastructure using mobile phone,
Wi-Fi Direct, or Bluetooth interfaces carried by volunteers operating in DTN mode [1–6,
19]. Such networks can be used to collect and transmit post-disaster shelter needs to the
EOC on a peer-to-peer basis through the volunteers carrying smartphones. The proposed
BlockCent scheme, running on the blockchain-DTN integrated environment, helps in fast
transmission of relief shelter needs to the EOC by enforcing cooperation among the for-
warding nodes. This leads to efficient disaster relief, even in an intermittently connected
environment.
The rest of this chapter is organized as follows. Section 10.2 summarizes related work
in this field. In Section 10.3, we provide an overview of the rudimentary elements of the
blockchain system. Section 10.4 describes the system model of the proposed scheme. We
elaborate the mechanism for integrating blockchain with DTNs in Section 10.5 and illustrate
the blockchain-based incentive scheme in Section 10.6. A detailed security analysis is given in
Section 10.7. Section 10.8 presents the experimental results. We conclude the chapter with a
direction towards future work in Section 10.9.

10.2 LITERATURE REVIEW

Incentive schemes for DTN applications can be classified into three categories: reputation-
based schemes, barter-based schemes, and credit-based schemes. Since the proposed work
deals with incentivizing DTN nodes for cooperation, we review a few prominent works on
credit-based incentive schemes for DTNs only.
Zhu et al. [9] propose SMART, a secure, multilayer, credit-based incentive scheme to stimu-
late bundle forwarding cooperation among DTN nodes. To address the selfishness problem in
DTNs, Lu et al. [10] propose a practical incentive protocol, called Pi, such that when a source
node sends a bundle message, it also attaches some incentive on the bundle, which is not only
attractive but also fair to all participating DTN nodes. Shevade et al. [11] propose the use of
190  Opportunistic Networks

pair-wise tit-for-tat (TFT) as a simple, robust, and practical incentive mechanism for DTNs.
Authors also develop an incentive-aware routing protocol that allows selfish nodes to maxi-
mize their own performance while conforming to TFT constraints. Chen et al. [12] propose
MobiCent, a credit-based incentive system for DTN. While MobiCent allows the underlying
DTN routing protocol to discover the most efficient paths, it is also incentive compatible.
Ning et al. [13] present effective schemes to estimate the expected credit reward, and they
formulate nodal communication in a DTN as a two-person cooperative game, whose solu-
tion is found by using the Nash Theorem. MobiID, a user-centric and social-aware reputation
based incentive scheme for DTNs, proposed by Wei et al. [14], allows a node to manage and
demonstrate its reputation.
All of these schemes either (i) rely on central trusted authorities that do not exist in a
post disaster communication network, or (ii) use no explicit virtual digital currency system.
These issues are addressed in the works proposed by He et al. [8], where a blockchain-based
incentive mechanism for distributed P2P applications uses Bitcoin to incentivize forwarders,
and Park et al. [7], where a Bitcoin-based incentive scheme incentivizes forwarders in vehic-
ular DTNs. However, both [7] and [8] assume that all forwarders are equipped with Internet
connectivity and directly access the Blockchain network. Such assumption is impracticable
in several intermittently connected network environments, such as the post-disaster com-
munication network, where forwarder nodes are DTN-enabled smartphones communicat-
ing over Wi-Fi Direct but having no Internet connectivity. The proposed blockchain-based
incentive scheme is inspired by the existing research but is very different. Major contribu-
tions of this work are:

• A mechanism that enables usage of blockchain technology in an intermittently con-

nected network such as DTN
• A fully decentralized blockchain-based node incentivizing scheme for DTNs
• without relying on central trusted authorities
• without requiring end-to-end Internet connectivity
• A novel reward model to make the incentivizing process more rational
• Extensive security analysis to justify security of the proposed incentivizing scheme
• Exhaustive simulations on the ONE simulator and the Ethereum platform substantiate
the efficiency of the incentivizing scheme

10.3 RUDIMENTARY ELEMENTS OF BLOCKCHAIN

Blockchain is a system in which a record of transactions made in Bitcoin or any other crypto-
currency are maintained across several computers that are linked in a peer-to-peer network.
It is a distributed database at the same time. A blockchain is constantly growing as new sets
of records/blocks are added to the chain [7, 18]. The overview of blockchain preliminaries
relevant to our proposal is given next.

10.3.1 Blocks
In a blockchain, transactions are grouped into blocks so that they can be efficiently verified.
Once a block is verified, they are added to the previous block. In a blockchain, each block is
chained to its previous block through the use of a cryptographic hash. A change to a block forces
a recalculation of all subsequent blocks, which requires enormous computation power. This
makes the blockchain immutable, a key feature of crypto currencies like Bitcoin and Ethereum.
 Blockchain Leveraged Node Incentivization  191

10.3.2 Transaction
A blockchain transaction can be defined as a small unit of task that is stored in a block.
A typical transaction [7] from a sender to a recipient is shown in Figure 10.1. A transaction
has a unique identifier (TX) and consists of a set of inputs and outputs. Inputs to a transac-
tion include previous transaction identifier (TX') and signature of the user. Outputs from a
transaction include the amount transferred to a recipient and the recipient’s public key. To
authorize spending the coins in the transaction input, the user should present its signature
for the transaction and corresponding public key. Output of a transaction is used as input of
the next transaction for verification. Transaction verification is distributed to P2P network
nodes, and only the valid transactions are recorded.

10.3.3 Mining
Mining involves creating a hash of a block of transactions that cannot be easily forged,
protecting the integrity of the entire blockchain without the need for a central system. Some
computers storing the entire blockchain are called miners. When a transaction is made over a
blockchain, the transaction information is thus put on a block and gets recorded. All the min-
ers crack a cryptographic riddle so as to locate the correct cryptographic hash. When a miner
creates the hash, it is added to the blockchain and must be confirmed by different nodes on
the system in a cycle known as consensus. At the point when a miner effectively confirms the
hash, it is given a reward [18].

10.3.4 Ethereum and Smart Contracts

Ethereum [18] is an open-source, public, blockchain-based distributed computing platform.
Ether is the cryptocurrency generated by the Ethereum platform as a reward to mining nodes
for computations performed and is the only currency accepted in the payment of transac-
tion fees. Ethereum provides a decentralized virtual machine, the Ethereum Virtual Machine
(EVM), which can execute scripts using an international network of public nodes. “Gas” is
the fee, or pricing value, required to successfully conduct a transaction or execute a contract
on the Ethereum blockchain platform.
A smart contract is a piece of code that runs on Ethereum to execute credible transactions
without any third parties. It is a self-executing contract between network participants, with-
out the need for traditional legal contracts. The contracts are high-level applications that are
compiled down to EVM bytecode and deployed to the Ethereum blockchain for execution.

Figure 10.1  Overview of blockchain transaction structure and spending [7].

192  Opportunistic Networks

An Ethereum user creates a contract and pushes the data to that contract so that it can
execute the desired command.

10.4 INTEGRATION OF BLOCKCHAIN WITH DTNS

In this section, we propose a mechanism for integrating blockchain technology with DTNs.
The major challenges of such integration are establishing the connection between DTN
and the Ethereum server, converting DTN messages to blockchain-compatible blocks, and
deploying blocks to the blockchain network.
We develop an application program interface (API) and an application binary interface
(ABI) that play the pivotal role in deploying the blocks containing DTN messages, in the form
of smart contracts on the Ethereum network. Our specific contributions are:

• Developing an API for establishing the connection between a DTN and the Ethereum
server
• Developing an ABI for converting the blocks, containing DTN messages, to Ethereum
blocks using Solidity contracts
• Deploying the converted blocks on the blockchain network using JSON-RPC protocol

In this work we use programming languages like Java and Solidity; frameworks like Spring
Boot, Spring Rest, Spring MVC, Spring Data, and Spring Security; and technologies like
Maven-Building tool, Metamask-Crypto wallet, and gateway to blockchain wallet and
Rinkeby-Ethereum Test Network. Initially, a Maven & Spring Boot project is created which
adds all dependencies (Spring Rest, MVC, Spring Data, Spring Security, JSON-RPC, ABI,
Lombok, Web3J, Slf4j, etc.) and creates the pom.xml and application.properties files. After
creating the project successfully, we develop the API that works as the bridge between the
local blockchain data and the Ethereum network using the Rest API Controller. The Rest API
Controller creates a link (“/accept-local blocks-from-dropbox”) with support of Get Request
and Post Request. The DropBox can send data to the API control using this API.
After receiving the blocks, containing DTN messages, the API control frames the blocks as
Pojos (Plain old Java Objects). All Pojos are then converted into Solidity contracts using the
developed ABI. These Solidity contracts are framed as Ethereum blocks. Thus, DTN blocks
are converted into Ethereum blocks. These blocks are deployed to the Ethereum server in real
time, using the JSON-RPC protocol. Figure 10.2 elucidates the functioning of the developed
API control.

Figure 10.2  The designed API control.

 Blockchain Leveraged Node Incentivization  193

After the blocks are deployed on the Ethereum server, the server mines all blocks using its
own miners and pays a certain amount of gas to the miners as payment for mining the blocks.
After every successful mining, all DTN blocks are deployed into the blockchain network,
which is completely immutable.

10.5 SYSTEM MODEL

In this section, we provide the system model, consisting of the network architecture and
security issues, pertaining to the disaster management use case of the proposed BlockCent
scheme.

10.5.1 Network Architecture
During and after disasters, victims normally take shelter in nearby safe areas. An EOC is
set up at nearby resource-rich areas [1–5, 19]. Our proposed scheme maps the earlier sce-
nario where the smartphones carried by the volunteers communicate over a DTN and create
a post-disaster communication network. In this scheme, the entire disaster-affected area is
virtually divided into a number of non-intersecting zones, having many shelters. Network
architecture used in our scheme, illustrated in Figure 10.3, consists of six major components,
whose functions are described next.

Figure 10.3  Network architecture of the proposed scheme.

194  Opportunistic Networks

Shelter node: Each shelter has a shelter node (for example, a laptop having Wi-Fi Direct
and LTE/5G Internet connection). Emergency managers in a shelter upload the shelter
requirements to the shelter node, which transmits this data to the volunteers working
in and around it.
Control node: The EOC has a control node (e.g. a workstation fitted with an LTE/5G
Internet link and Wi-Fi Direct) that gathers situational information and distributes
resources to the remote shelters.
Forwarder node: Volunteers carrying Wi-Fi Direct-equipped smartphones are forwarder
nodes that travel around the disaster-stricken region, collect shelter node situational
messages, communicate these messages with each other through the PRoPHET [18]
DTN routing protocol, and eventually transmit them to the control node.
Observer node: Disaster management agents equipped with Wi-Fi Direct smart devices
are observer nodes [20] that travel at a pace greater than that of the volunteers around
the disaster-stricken region. The zone does have a designated observer node which col-
lects and transmits to the nearest DropBox all transactions created by forwarder nodes
therein.
DropBox: Stationary relay nodes (e.g., smartphone, laptop having LTE/5G Internet con-
nectivity, possessing sufficient storage and computation capacity) that receive and store
transactions submitted by the observer nodes passing by it and forward them to the API
control, running on a cloud platform, for further processing. DropBoxes are deployed
at high-priority locations, based on the uDBdep scheme [21] to maximize data transfer.
Ethereum server: The Ethereum server runs the open-source, public, blockchain-based dis-
tributed computing system. It mines blocks and uploads them to the global blockchain.

Each of these components runs different modules of the proposed scheme to accomplish mes-
sage forwarding and appropriate incentivizing.

10.5.2 Security Model
In this work, we assume the control node to be trusted, and all security issues, described next,
arise from selfish shelter nodes, forwarder nodes, and observer nodes.

• A shelter node may refuse to pay back the cooperating forwarder nodes and observer
node in its zone when the message finally reaches the control node.
• A forwarder node may not:
• forward a message;
• send acknowledgment to the previous forwarder node.
• An observer node may pose as a forwarder node and claim incentives for forwarding
messages.

These selfish behaviors can be further complicated by the collusion of two or more nodes.
However, such collusion-based security issues are out of the scope of the current work.

10.6 BLOCKCENT: BLOCKCHAIN-BASED NODE INCENTIVIZING SCHEME

In this section, we present BlockCent, a blockchain-based node incentivizing framework for

a smartphone-powered DTN that runs on the integrated blockchain-DTN setting. In this
scheme, when transmitting a message to the control node through one or more forwarder
nodes, a shelter node belonging to a specific zone gives a certain amount of Ethereum as a
 Blockchain Leveraged Node Incentivization  195

reward to all cooperative forwarder nodes that assist in transmitting the message and the
observer node inside it. Each intermediate forwarder node collects from the next-hop for-
warder a digitally signed acknowledgment (trust-token) to which it forwards the message as
a sign of cooperation. If and only if it has a digitally signed acknowledgment from its succes-
sor, then a forwarder node is called cooperative. It should be remembered that each reward
is actually a promise, and the forwarder nodes can reclaim rewards only after the shelter
node recognizes that the message is successfully transmitted to the control node. This helps
in thwarting the forwarder nodes from taking refuge in “dine and dash” behavior.

10.6.1 Setup Phase
In our scheme, we assume N forwarder nodes, K shelter nodes, P observer nodes (one per
zone) and one control node. In the setup phase, each shelter node is assigned a shelter id SNj,
j = 1, 2, …, K,and each forwarder node a forwarder id FNi, i = 1, 2, …, N, by the control
node. Each shelter node and each forwarder node is also assigned a private and public key
pair (PRSNj, PUSNj), j = 1, 2, …, k and (PRFNi, PUFNi), i = 1, 2, …, N, respectively, by the control
node. The control node has its own private and public key pair (PRCN, PUCN). All public keys
are distributed in the network. Whenever a new volunteer wants to join relief operations,
he/she has to register with EOC. On registration, the message transmission and incentive
redemption modules of the proposed blockchain-based incentive application is installed on
the volunteer’s smartphone along with necessary details like private and public key pair of its
own and public key of the other stakeholders.

10.6.2 Incentivizing Process
A reward is known as a transaction in the proposed scheme, and a shelter node uses the
blockchain method to enable cooperative forwarder nodes with sufficient quantities of
Ethereum. Three modules comprise the incentivizing process. First, a certain deposit that is
used to facilitate the cooperative forwarder nodes is made by the shelter node and generates
a transaction. Second, the shelter node transmits the message through one or more forwarder
nodes to the control node. Finally, the forwarder nodes that cooperate in forwarding messages
redeem their incentives.
To illustrate the incentivizing process, we assume that a shelter node SNj in zone Z sends
message M to the control node CN through the cooperative forwarder nodes FN1, FN2, and
FN3. The observer node ONZ moves around Z for collecting transactions generated by FN1,
FN2, and FN3 and broadcast them to the blockchain network for necessary validation and
inclusion in the blockchain. As mentioned in Section 10.5.1, the forwarder nodes exchange
messages using the PRoPHET protocol [17]. This protocol uses history of previous encoun-
ters and transitive property to estimate delivery predictability (DP), P(A, B), of each node A
for all known destinations B. P(A, B) depicts the probability of A meeting B. As nodes meet,
they exchange DP tables and update their own DP table according to certain protocols.

10.6.2.1 Transaction Creation
SNj generates two nonce values N1and N2; N1 is to be used for proving that FN1, the next
hop forwarder node receives the message correctly and N2 is to be used for proving that CN
receives the message successfully. SNj publishes EPUSNj(N1) and EPUSNj(N2) to the blockchain
network over the Internet. SNj then makes a deposit to commit that it will give incentives to
FN1 if FN1 cooperates to transmit the message to CN. SNj constructs a time-locked deposit
Deposit
transaction TXSN j
and broadcasts it to the blockchain network. The time-lock condition
196  Opportunistic Networks

restricts SNj from withdrawing the deposited Ethereum earlier than the locking period. The
locking period could be the maximum time required by the miners to verify the transaction
based on information provided by FN1, FN2, FN3, and CN.

10.6.2.2 Message Transmission

SNj composes and sends the message bundle msg1 ≔ {M‖EPUCN(N2)‖SigPRSNj(Hash(M)‖EPUCN

(N2))‖SigPRSNj(N1)} to its next-hop forwarder node FN1 over DTN, using the DTN rout-
Reward
ing protocol. SNj constructs a reward transaction TXSN j → FN1
and broadcasts it to the block-
chain network for validation along with delivery predictability P(FN1, CN), which depicts the
probability of FN1 meeting CN, final destination of M.
On receiving msg1, FN1 extracts N1 and saves it to be used for incentive redemption. It
also verifies the signature {SigPRSNj(Hash(M)‖EPUCN(N2))} and sends SNj a digitally signed and
encrypted acknowledgment EPUSNj(SigPRFN (ACKFN1)), encrypted with SNj’s public key. When
1
FN1 opportunistically meets FN2, it forwards the message bundle msg2 ≔ {M‖EPUCN(N2)‖S
igPRSNj(Hash(M)‖EPUCN(N2))} to its next-hop forwarder node FN2 over DTN and constructs
Reward
a reward transaction TXFN 1 → FN 2
.Since FN1 is a normal forwarder node with less speed, it
Reward
carries TXFN1 → FN2 along with P(FN2, CN) until it meets the high-speed observer node ONZ
opportunistically. On receiving these values, ONZ transfers them to the DropBox, which in
turn transmits them to the API control, running on a cloud server, over the Internet.
A similar process is executed between FN2 and its next-hop forwarder node FN3.
Finally, when FN3 reaches CN, it forwards the message bundle msg4 ≔ {M‖EPUCN(N2)
‖SigPRSNj(Hash(M)‖EPUCN(N2))} to CN. CN receives the message, verifies the signature
{SigPR (Hash(M)‖EPUCN(N2))},and sends FN3 a digitally signed acknowledgment EPUFN3(SigPRCN
SNj
(ACKCN)), encrypted by FN3’s public key.
Reward Reward Reward
Transactions TXSN j → FN1
, TXFN 1 → FN 2
, and TXFN 2 → FN3
are commitments that coins will be
transferred only after the miners have verified them.

10.6.2.3 Incentive Redemption
Once the message M is successfully delivered to CN, CN provides {EPUCN(N2), EPUCN(ACKCN)}
and SNj provides ESNj(ACKFN ) to the blockchain miners. The Ethereum server collects from
1
the API control the proofs of cooperation of FN1, FN2, and FN3 and submits them to the
blockchain miners. In particular, the Ethereum server collects:

{EPUFN1(N1), EPUFN (ACKFN2), PUSNj, PUFN2} from FN1,

1
{EPUFN (ACKFN2), EPUFN (ACKFN3), PUFN3} from FN2,
2 2
{EPUFN (ACKFN3), EPUFN (ACKCN), PUCN} from FN3.
3 3

The incentives could be redeemed only if the miners validate the transactions using these
values and the previously encrypted nonce values broadcast by SNj. The conditions for valid-
ity are as follows, all encryption functions being commutative:

• The encrypted nonce provided by CN can be verified with the encrypted nonce provided
by SNj as:

 
EPUCN EPUSN j  N 2   EPUSN j  EPUCN  N 2  
 Blockchain Leveraged Node Incentivization  197

• The encrypted nonce provided by FN1 can be verified with the encrypted nonce pro-
vided by SNj as:

 
EPUFN 1 EPUSN j  N1   EPUSN j  EPUFN1  N1  

• There is a delivery path from SNj to CN verified by the acknowledgments:

ACKFN1 → ACKFN 2 → ACKFN 3 → ACKCN

• Digitally signed acknowledgments provided by FN1, FN2, and FN3 can be verified as:

  
EPUFN1 EPUFN2  ACKFN 2   EPUFN2 EPUFN1  ACKFN 2  

EPUFN2 E PUFN3  ACKFN 3   E
PUFN3 E PUFN2  ACKFN  
3

EPUFN3  EPUCN  ACKCN    EPUCN  EPUFN3  ACKCN  

Incentives are rewarded to cooperative forwarder nodes and the observer node by the
blockchain network through the Ethereum server, API control, and the DropBoxes. The
BlockCent scheme, running on the blockchain-DTN integrated environment is illus-
trated in Figure 10.4.

10.6.3 Reward Model
A popular rewarding method in incentive schemes is paying per message, where, for a suc-
cessfully transmitted message, each of the n cooperative forwarder nodes receives x amount
of credit and the sender needs to pay x × n in total. However, in an opportunistic network
like ours, it is not possible to predict how many forwarder nodes will get involved in the
message transmission process. We therefore follow the approach of benefit sharing, where a
portion of the profit is shared by each of the n cooperative forwarder node, i.e. the amount
Deposit
deposited by the sender. Assuming that sender SNj has made the deposit transaction TXSN j

Figure 10.4  BlockCent scheme running on the blockchain-DTN integrated environment.

198  Opportunistic Networks

and deposited an amount of X at the beginning of transmission, and there exists n coopera-
tive forwarder nodes FN1, FN2, …, FNn that are involved, the reward is:

  , if U  ON Z

RU  
 X    P  FN ,CN  , if U  FN , i , 2, , n
 n i i

Incorporating delivery predictability P(FNi, CN) in the reward model makes the scheme more
reasonable. First, nodes having a higher DP with the control node are given higher rewards
to ensure faster message delivery. Second, in order to reduce the number of hops to optimize
their benefit, forwarder nodes try to forward messages to forwarders with the highest DP.
This in turn results in the delivery of quick messages.

10.7 SECURITY ANALYSIS

We provide a thorough security analysis of the proposed incentivizing scheme with respect to
the security issues mentioned in Section 10.5.2.

10.7.1 Shelter Node Refuses to Pay Back

A shelter node SNj makes a deposit at the beginning of message transmission, committing
Deposit
to pay back all stakeholders. SNj prepares a time-locked deposit transaction TXSN j
and
broadcasts to the blockchain network. The time-lock restricts SNj from withdrawing the
deposit before the expiration of the locking period and refuses to pay back. This guarantees
fairness to cooperating forwarders and the observer.

10.7.2 Forwarder Node Refuses to Forward Message

If a forwarder node FNi does not forward a message and to the next-hop forwarder node FNi + 1,
FNi will not receive the digitally signed acknowledgment EPUFN (SigPRFN (ACKFNi + 1)) from
i i+1
FNi + 1. Consequently, FNi is unable to provide EPUFNi(ACKFNi + 1) to the observer node ONZ at
the time of incentive redemption. Thus, FNi cannot get incentive without forwarding a message.

10.7.3 Forwarder Node Refuses to Send ACK

If a next-hop forwarder FNi + 1 refuses to send acknowledgment to its previous forwarder FNi,
FNi is unable to provide EPUFNi(ACKFNi + 1) to ONZ during incentive redemption. However, if
FNi + 1 provides EPUFNi + 1(ACKFNi + 2) to OZ, the miners will fail to verify the delivery path …
→ ACKFNi → ACKFNi + 1 → ACKFNi + 2 → … and will invalidate the transaction. Thus, FNi + 1
cannot get incentive in spite of forwarding the message.

10.7.4 Observer Node Poses as Forwarder Node

If an observer node ONZ intends to pose as a forwarder node FNi, it has to provide {EPUOz(A
CKONZ), EPUOz(ACKFNi + 1), PUONz} to the miners. However, ONZ has no ACKONZ to its credit,
nor can it decrypt ACKFNi + 1 with PRFNi to create EPUOz(ACKFNi + 1). Thus, ONZ cannot pose
as FNi and claim incentives for forwarding messages.
 Blockchain Leveraged Node Incentivization  199

10.8 EXPERIMENTAL RESULTS

In this section we evaluate the performance of the proposed BlockCent scheme through
extensive experiments conducted in a realistic simulation environment.

10.8.1 Simulation Environment
We use a real disaster scenario for setting up the simulation environment, based on the 2015
Nepal earthquake. The Google Map of water, food, shelter, and medical resources for the
Nepal earthquake [22], shown in Figure 10.5, marks the shelters and medical relief centers
set in Kathmandu and its adjoining districts like Nuwakot, Sindhupalchowk, and others. We
set up our simulation environment based on Figure 10.5.

10.8.2 Simulation Setup
The three main objectives of the experiments are (i) to observe how far BlockCent succeeds
in achieving major design goals, (ii) to evaluate network performance of BlockCent, and (iii)
to observe the overheads introduced by BlockCent for blockchain activities. We use the ONE
simulator [23] optimization tool for accomplishing the first two objectives and the Ethereum
[24] blockchain platform for the third.

10.8.2.1 ONE Setup
Our simulation setup in the ONE simulator [23] is based on information given by the map
[22] concerning the post-disaster relief operation carried out in the Kathmandu region of
Nepal. We set up 11 shelters covering about 100 sq. km across the disaster-hit city. We
set up one EOC as well. We consider 88 forwarder nodes, that is, mobile volunteers car-
rying smartphones. Such nodes share and relay situational messages from the shelters to
the EOC. The entire region affected by the earthquake is divided into five virtual areas. To
collect the transactions of the forwarder nodes in that zone, each zone has one observer
node. There are stationary shelter nodes and the control node. The forwarder nodes and
observers, however, are mobile nodes and adopt the Post Disaster Mobility (PDM) model

Figure 10.5  Google Map for the 2015 Nepal earthquake [22].
200  Opportunistic Networks

suggested in [25] by Uddin et al. A shelter node produces and transmits situational mes-
sages to the forwarder nodes at a rate of one per hour. We use real situational messages
originating from the WhatsApp chat log of an NGO’s “Doctors For You” volunteers [26],
who provided medical assistance during and after the earthquake. These messages are 120
bytes in size, containing the real emergency resource specifications at various shelters. In
our network, we consider cooperative forwarder nodes (CFN) and selfish forwarder nodes
(SFN). Cooperative forwarder nodes take part in the PRoPHET [27] routing protocol rout-
ing operation, while selfish forwarder nodes drop all messages sent to them (destined for
other nodes) that were supposed to be redirected to other forwarder nodes. The findings
presented here are the 20 independent run average. The relevant parameters used in simula-
tion are described in Table 10.1.

10.8.2.2 Ethereum Setup
We implement and evaluate a prototype of our proposed BlockCent scheme using Ethereum
[24] as the blockchain platform. Our entire code is approximately 500 lines, consisting of
smart contracts written in Solidity. The rationale behind using Ethereum is that it provides
a broader range of validation strategies compared to the other blockchain platforms, with
the power of smart contracts. We deployed our implementation on a private Ethereum
network consisting of one node, namely the DropBox running on Truffle v4.5. We have
used a single machine with Intel® Core™ CPU with 2.0 GHz, 8 GB of RAM, and 20 GB of
persistent storage to meet the minimum hardware requirement for running Ethereum. Also,
as a wallet for message transferring, we have used Geth v 1.7.0. An initial gas of 3,000,000
wei is used.

10.8.3 Simulation Metrics
Performance of the proposed BlockCent scheme is measured based on design metrics, network
performance metrics, and blockchain metrics.

Table 10.1  Parameters used for simulation on ONE.

Parameter Value

Simulation area 100 sq. km

No. of forwarder nodes 88
No. of shelters 11
Message generation rate 5 per hour
Message size 120 bytes
Message TTL 500 min
No. of zones 5
No. of observer nodes 5 (1 in each zone)
Forwarder node speed 0.5 m/s to 1.5 m/s
Observer node speed 1.5 m/s to 5.5 m/s
Transmission range 10 m
Movement model Post-disaster mobility model
Routing protocol PRoPHET
Simulation period 12 hours (43,200 sec)
 Blockchain Leveraged Node Incentivization  201

10.8.3.1 Design Metrics
Selfishness ratio: It is the fraction of the number of selfish forwarder nodes to the total
number of such nodes. It is defined as

No. of selfish forwarder  nodes

SelfishnessRatio 
Total no. of forwarder  nodes

This metric is used to evaluate the ability of the BlockCent scheme to convert selfish nodes
to cooperative nodes.
Average reward: It is the average of the rewards obtained by all forwarder nodes for
forwarding all messages. It is defined as

AverageReward 
 RFN
all forwarder  nodes

No. of forwarder  nodes

where RFN is the total reward obtained during a simulation period.

10.8.3.2 Network Performance Metrics

Delivery ratio [5]: It is the fraction of the messages delivered to destination nodes to those
created by source nodes. It is defined as

No. of msgs.delivered
DeliveryRatio =
No. of msgs.created

Average delay [5]: This indicates the average time taken by the messages from sources to
destinations, including buffer delays, queuing delays, retransmission delays, and propa-
gation time. It is defined as

AverageDelay 
 forallMD
TD  TS 
No. of msgs.delivered

where TD is the time when message reaches its destination and TS is the time when
the message was created at the source.
Overhead ratio [5]: Indicates the ratio of the number of control messages (including route
request/reply/update/error packets) to the number of data messages. Defined as

No. of control packets relayed

OverheadRatio =
No. of data packets relayed

10.8.3.3 Blockchain Metrics
Processing time (tp) [28]: It is the amount of time a user has to wait, after pressing the
“send” transaction button (t1), to see their transaction appear on the blockchain (t2) and is
defined as

t p  t2  t1
202  Opportunistic Networks

Gas consumption (G) [29]: It is the gas consumed in processing transactions (tx = 1, 2, …, n)
for the current block and is defined as

G GasConsumption
t x 1
tx

10.8.4 Results and Discussion

In this section, we discuss the results of the experiments conducted for evaluating the
BlockCent scheme in terms of achieving major design goals, network performance and block-
chain overhead.

10.8.4.1 Achieving Major Design Goals

Two sets of experiments are conducted for evaluating the performance of the BlockCent
scheme by measuring the extent of attaining design objectives.
The first set of experiments measure the selfishness ratio against time, shown in Figure
10.6(a), where time varies from 1 to 12 hours. We start with 50% selfish forwarder nodes and
observe that the percentage of selfish forwarder nodes decrease with time. As more and more
cooperative nodes are rewarded for forwarding messages, selfish nodes are encouraged to
cooperate and participate in forwarding activities. Thus, the Block Cent scheme successfully

(a)

(b)

Figure 10.6  (a) Selfishness ratio with time. (b) Average reward with no. of messages per hour.
 Blockchain Leveraged Node Incentivization  203

converts selfish nodes to cooperative nodes. The graph in the inset depicts the increase in
selfishness ratio when the PRoPHET routing protocol is used without the BlockCent scheme.
The reward model used in the BlockCent scheme, described in Section 10.6.3, includes
delivery predictability of forwarder nodes with the control node. Thus, messages are always
forwarded through forwarder nodes having higher delivery predictability, which in turn
reduces the number of hops and maximizes reward for cooperating nodes. The second set
of experiments measure the average reward of the forwarder nodes against the number of
messages per hour. Figure 10.6(b) shows that the average reward increases with increasing
number of messages per hour using this reward model. The graph in the inset depicts that
average reward does not increase substantially when delivery predictability is not included
in the reward model, which is justified as most eligible forwarders are not chosen and the
number of hops is increased.

10.8.4.2 Evaluation of Network Performance

We compare our BlockCent scheme with three other competing schemes, namely SMART
[9], MobiCent [13], and MobiID [14], in terms of network performance metrics. To evaluate
network performance, three sets of experiments are conducted.
In the first set of experiments, the performance of BlockCent is compared with other com-
peting schemes in terms of delivery ratio with varying numbers of selfish forwarder nodes in
the network. It is observed from Figure 10.7(a) that delivery ratio reduces with increasing
number of selfish nodes for all schemes. However, BlockCent performs the best due to the
appropriate reward model that incentivizes cooperative nodes for forwarding messages, thus
boosting delivery performance. SMART performs second best, by having 12% less delivery
ratio than BlockCent.

(a) (b)

(c)

Figure 10.7  (a) Delivery ratio with different levels of selfishness. (b) Average delay with different levels of self-
ishness. (c) Overhead ratio with different levels of selfishness.
204  Opportunistic Networks

In the second set of experiments, the performance of BlockCent is compared with other
schemes in terms of average delay (in seconds) with varying number of selfish nodes in
the network. Figure 10.7(b) indicates that BlockCent performs best, followed by MobiID.
Including delivery predictability into the reward model encourages forwarder nodes to for-
ward messages to forwarders with highest DP to reduce the number of hops for maximizing
their profit. This in turn results in fast message delivery. The gain in average delay over the
SMART scheme is 22%.
The third set of experiments measures overhead ratio with varying number of misbehaving
nodes for all three techniques. As shown in Figure 10.7(c), overhead ratio of BlockCent
is high in comparison to the other schemes. This increase can be attributed to the control
packets, consisting of reward TX, DPs, and incentive redemption values exchanged between
the forwarder nodes and the observer nodes used for reward redemption. The MobiCent
scheme performs best followed by the MobiID and SMART schemes. However, BlockCent
has only 15% greater overhead than MobiCent, which is not alarming.
It is evident from the preceding discussion that BlockCent maximizes network performances
in terms of delivery ratio and average delay while maintaining a tolerable overhead ratio.

10.8.4.3 Overheads Introduced for Blockchain Activities

Two sets of experiments are conducted to observe the overheads introduced by the BlockCent
scheme for blockchain activities.
In the first set of experiments, the impact of varying number of shelters and zones on the
processing time of BlockCent is measured, as depicted in Figure 10.8(a).We observe that tp
increases with the increase in the number of shelters due to the higher number of transactions
to be verified. For example, with 4 zones, tp with 8 shelters is 22.85% and 15.82% higher
than the tpwith 4 and 6 shelters, respectively. It is also observed that tp does not increase sig-
nificantly with increasing number of zones, as a slight increase in computation is involve in
this case. Only 2.5% is the average increase in tp with increasing number of zones.
In the second set of experiments, the impact of varying numbers of shelters and zones on
gas consumption of BlockCent is measured, as shown in Figure 10.8(b). We observe that gas
consumption increases with increasing number of shelters, which incurs higher number of
transactions to be verified. For example, with 4 zones, G with 8 shelters is 10.2% and 5.75%
higher than the G with 4 and 6 shelters respectively.However, the rate of increase in the case
of zones is not so significant. Only 2.95% is the average increase in G with increasing num-
ber of zones.

10.9 CONCLUSION

This chapter proposed BlockCent, a blockchain-based node incentivizing scheme for a DTN
enabled smartphone leveraged communication network that uses Ethereum to incentivize
nodes for cooperation, built on top of a DTN-blockchain integrated environment. This envi-
ronment develops an alternative way of broadcasting and authorizing blockchain transac-
tions without having to rely on Internet connectivity. It uses the store-carry-forward feature
of DTN nodes in conjunction with stationary relay nodes (e.g., smartphone, laptop), called
DropBoxes, to transmit and authorize transactions, hence enabling the usage of blockchain
in intermittently connected network environments. The incentivizing scheme is fully decen-
tralized and does not rely on any central trusted authority. A novel reward model is used
to make the incentivizing process fair and rational. The proposed mechanism is applied to
the disaster management use case, based on the Ethereum platform using smart contracts in
 Blockchain Leveraged Node Incentivization  205

(a)

(b)

Figure 10.8  (a) Processing time. (b) Gas consumption.

Solidity. It successfully integrates blockchain with the DTN environment towards improving
disaster management services in absence of end-to-end to the Internet.
As future work, we intend to make the proposed scheme resilient to collusion attacks by
introducing anonymity of the forwarder nodes. One-time public key techniques, such as
CryptoNote [30], can be used for this purpose.

REFERENCES

[1] A. Verma, M. Singh, K. K. Pattanaik, and B. K. Singh (2019) Future networks inspired by oppor-
tunistic network, Opportunistic Networks: Mobility Models, Protocols, Security & Privacy, CRC
Press (Taylor & Francis Group), chapter 12, pp. 229–246, 2019. doi:10.1201/9780429453434
[2] S. Basu, A. Biswas, S. Roy, and S. DasBit (2018) Wise-PRoPHET: A watchdog supervised
PRoPHET for reliable dissemination of post disaster situational information over smartphone
based DTN. J Netw Comput Appl 109(2018), 11–23.
[3] S. Basu and S. Roy (2014) A global reputation estimation and analysis technique for detection of
malicious nodes in a post-disaster communication environment. In Proceedings of AIMoC 2014.
[4] S. Basu and S. Roy (2014) A group-based multilayer encryption scheme for secure dissemination
of post-disaster situational data using peer-to-peer delay tolerant network. In Proceedings of
ICACCI 2014, pp. 1566–1572.
[5] C. Chakrabarti and S. Basu (2019) A blockchain based incentive scheme for post disaster
opportunistic communication over DTN. In Proceedings of the 20th International Conference
on Distributed Computing and Networking, January 2019, pp. 385–388, https://doi.
org/10.1145/3288599.3295584.
206  Opportunistic Networks

[6] J. Ren, Y. Zhang, K. Zhang, and X. S. Shen (2015) Exploiting mobile crowdsourcing for perva-
sive cloud services: Challenges and solutions. IEEE Commun Mag 53(3), 98–105.
[7] Y. Park, C. Sur and K. H. Rhee (2018) A secure incentive scheme for vehicular delay tolerant
networks using cryptocurrency. Secur Commun Netw 2018, 1–13.
[8] Y. He et al. (2018) A blockchain based truthful incentive mechanism for distributed P2P applica-
tions. IEEE Access 2018(6), 27324–27335.
[9] H. Zhu et al. (2009) SMART: A secure multilayer credit-based incentive scheme for delay-toler-
ant networks. IEEE Trans. Veh. Tech. 58(8), 4628–4639.
[10] R. Lu, X. Lin, H. Zhu, X. Shen, and B. Preiss (2010) “Pi: A practical incentive protocol for delay
tolerant networks.” IEEE TWC.
[11] U. Shevade, H. H. Song, L. Qiu, and Y. Zhang (2008) Incentive-aware routing in DTNs. In
Proceedings of ICNP.
[12] B. B. Chen and M. C. Chan (2010) MobiCent: A credit-based incentive system for disruption
tolerant network. In Proceedings of INFOCOM.
[13] T. Ning, Z. Yang, X. Xie, and H. Wu (2011) Incentive-aware data dissemination in delay-tolerant
mobile networks. In Proceedings of SECON 2011.
[14] L. Wei, H. Zhu, Z. Cao, and X. S. Shen (2011) MobiId: A user centric and social-aware reputa-
tion based incentive scheme for delay/disruption tolerant networks. In Proceedings of ADHOC-
NOW 2011.
[15] S. Nakamoto (2008) Bitcoin, A peer-to-peer electronic cash system. https://bitcoin.org/bitcoin.
pdf.
[16] M. Crosby et al. (2015) Blockchain technology: Beyond Bitcoin. Sutardja Center for
Entrepreneurship & Technology Technical Report, Berkeley University of California 2015, 1–35.
[17] A. M. Antonopoulous (2017) Mastering Bitcoin – programming the open blockchain. O’Reilly
Media 2017.
[18] Understanding blockchain: A beginners guide to Ethereum Smart Contract Programming,
https://www.codemag.com/Article/1805061/Understanding-Blockchain-A-Beginners-Guide-to-
Ethereum-Smart-Contract-Programming (Accessed 9 July 2020).
[19] S. Basu, S. Bhattacharjee, S. Roy, and S. Bandyopadhyay (2015) SAGE-PRoPHET: A security
aided and group encounter based PRoPHET routing protocol for dissemination of post disaster
situational data. In Proceedings of ICDCN 2015, Article No. 20.
[20] C. Chakrabarti, S. Roy, and S. Basu (2018) Intention aware misbehavior detection for post-
disaster opportunistic communication over peer-to-peer DTN. Peer-to-Peer Networking and
Applications 2018, pp. 1–19.
[21] N. Das, S. Basu, and S. Das (2020) Efficient DropBox deployment towards improving post
disaster information exchange in a smart city. In: ACM Transactions on Spatial Algorithms and
Systems, 6(2), pp. 1–18.
[22] Map of water, food, shelter and medical resources. https://www.google.com/maps/d/viewer?mid
=1Iv7GILViqyJAFn5o5hi1 F2Fg8mc&hl=en_US. 2015.
[23] A. Keranen, J. Ott, and T. Karkkainen (2009) The ONE simulator for DTN protocol evaluation.
In Proceedings of SIMUTools, Rome, Italy, March 2–6, 2009.
[24] G. Wood (2014) Ethereum: A secure decentralized generalized transaction ledger. Ethereum
Project Yellow Paper, 151, pp. 1–32.
[25] M. Y. S. Uddin, D. M. Nicol, T. F. Abdelzaher, and R. H. Kravets (2009) A post-disaster mobil-
ity model for delay tolerant networking. In Proceedings of WSC, Texas, USA, December 13–16,
2009.
[26] Doctors For You. http://doctorsforyou.org.
[27] A. Lindgren, A. Doria, E. Davies, and S. Grasic (2011) Probabilistic routing protocol for inter-
mittently connected networks. draft-lindgren-dtnrg-prophet-09.txt, 2011.
[28] Regio A. Michelin et al. (2018). SpeedyChain: A framework for decoupling data from blockchain
for smart cities. In Proceedings of 15th EAI International Conference on Mobile and Ubiquitous
Systems: Computing, Networking and Services, ACM, 145–154.
[29] Gas usage, https://www.investopedia.com/terms/g/gas-ethereum.asp (Accessed 28 June 2020).
[30] N. van Saberhagen (2013) Cryptonote v2.0 2013. https://cryptonote.org/whitepaper.pdf.
Chapter 11

Evaluation of Energy Efficiency of

Opportunistic Network Routing Protocols
Md. Khalid Mahbub Khan, Muhammad Sajjadur Rahim and
Abu Zafor Md. Touhidul Islam
University of Rajshahi, Rajshahi, Bangladesh

CONTENTS

11.1 Introduction and Related Works.............................................................................. 207

11.2 Routing Protocols for OppNet................................................................................. 209
11.2.1 Replication-Based Protocols....................................................................... 209
11.2.1.1 Epidemic..................................................................................... 210
11.2.1.2  Spray and Wait............................................................................ 210
11.2.1.3  Spray and Focus.......................................................................... 210
11.2.2 Probabilistic Protocols................................................................................ 211
11.2.2.1 PRoPHET ................................................................................... 211
11.2.3 Social-Based Protocols................................................................................ 212
11.2.3.1  Bubble Rap................................................................................. 212
11.2.3.2 SCORP....................................................................................... 212
11.2.3.3 dLife........................................................................................... 213
11.2.3.4 dLifeComm................................................................................. 213
11.3 Simulation Environments......................................................................................... 213
11.3.1 Simulation Tool..........................................................................................213
11.3.2 Simulation Scenario Settings....................................................................... 213
11.4 Performance Metrics ...............................................................................................215
11.4.1 Average Remaining Energy.........................................................................215
11.4.2 Delivery Ratio............................................................................................ 215
11.4.3 Average Latency......................................................................................... 215
11.4.4 Overhead Ratio.......................................................................................... 216
11.5 Results and Discussion............................................................................................. 216
11.5.1 Impact of Node Density.............................................................................. 216
11.5.2 Impact of TTL............................................................................................ 219
11.6 Conclusions............................................................................................................. 221
Acknowledgments............................................................................................................ 222
References........................................................................................................................ 222

11.1 INTRODUCTION AND RELATED WORKS

The opportunistic network (OppNet) is an enhanced version of the multi-hop ad hoc net-
work since the application fields of this network have been increased. This type of network-
ing scenario was proposed for military purposes. To improve the success rate of data delivery
on the battlefield among different military units was the primary issue. But in present days,
OppNet’s applications range has been broadened in civil aspects since the innovations of
many smart communicating devices, and people also want connectedness during their daily
207
208  Opportunistic Networks

activities [1]. The absence of a dedicated contact link from source to destination is the key
feature of OppNet. In this case, the performance of the communication link is variable in
either extreme. Conventional TCP/IP protocol will fail to perform efficiently as an end-to-end
transmission path may exist for a short or uncertain period in OppNet. Moreover, transmis-
sion delays as well as the error rate are also higher. So, node mobility is exploited here to
carry on legitimate data transmission in such cases, and mobile nodes are responsible for the
delivery of data [2]. Here, nodes select the next relay for data forwarding based on the “store-
carry-forward” manner. In this fashion, nodes can temporarily store data, and carry it for a
time by adapting the advantages of both node mobility and opportunistic contacts. Besides,
due to the mobility of node, OppNet becomes a heterogeneous network [3].
To identify the proper relay node for successful data delivery with reduced latency is the
major issue in OppNet. Regarding this issue, several forwarding strategies have been pro-
posed for OppNet from various points of view. The replication-based routing approach was
the first-generation protocols proposed for OppNet. Epidemic [4] was the first replication-
based protocol. Later, many improved and sophisticated protocols, viz. Spray and Wait [5],
Spray and Focus [6], etc. were proposed within this generation. Then the probabilistic con-
cept emerged in data forwarding. The PRoPHET [7] protocol is one of the examples of such
a genre. After that, social-based forwarding strategies were developed for improved data
forwarding in OppNet. Several social characteristics are considered to route data in such a
generation of protocol. SCORP [8], dLife [9], dLifeComm [9], Bubble Rap [10], etc. belong
to this category. Moreover, research has been going on in OppNet to find out the most effi-
cient routing solutions.
However, nodes in OppNet are mostly lithium-ion-battery driven. So, they are energy-
constrained. Nodes have to perform several activities such as transmitting and receiving mes-
sages, scanning neighboring nodes, etc., within their limited lifetime. A significant amount
of energy is also invested due to the mobility of nodes. So, the energy consumption of nodes
can be regarded as a vital aspect for OppNet. Furthermore, in practice, it is not easy, and
sometimes not possible, to recharge the battery of a node. If the energy of nodes is consumed
completely, then the temporal contact link between source and destination would be discon-
nected. Finally, data would not be delivered successfully to the destination [11]. Besides, the
battery technology has not sufficiently improved to keep pace with the increasing require-
ment of energy. Early draining energy is one of the major deficiencies for mobile devices [12].
So, energy should be utilized efficiently for better networking performance as well as success-
ful data delivery. Thus, the routing strategy should be energy-efficient.
Energy consumption with relevant performance evaluation of different routing protocols
has been considered in many pieces of research in OppNet scenarios previously. In [13–17],
the energy consumptions of different protocols, including some conventional OppNet pro-
tocols and a few social-based protocols, have been investigated with their comparative
performance analysis. These investigations have been performed on behalf of several met-
rics such as average remaining energy, count of dead nodes, delivery ratio, average latency,
overhead ratio, etc. In the aforementioned research works, networking scenarios have been
varied after considering some networking aspects such as node density, buffer size, mes-
sage generation rate, message TTL, networking area, etc. A performance evaluation of two
social-based protocols, SCORP and dLife, has been presented against the replication-based
and probabilistic protocols Epidemic, Spray and Wait, and PRoPHET in [18]. Moreover, in
[11], the authors provided a detailed study of energy-aware, social-based protocols. But the
experimental analysis has not exhibited there. A brief study about energy consumption and
comparative performance evaluation of six distinct conventional protocols in the OppNet
environment except considering any social-based protocol is provided in [19]. Here, also, the
impact of node density and TTL has been reflected. A perfect routing protocol in the OppNet
 Evaluation of Energy Efficiency  209

scenario must have a higher delivery ratio and minimum energy consumption, latency, and
overhead ratio. To choose the best forwarding strategy among all that exist, it is necessary
to know their comparative performance in the OppNet environment. This chapter provides
an extended study relative to energy efficiency as well as modified performance evaluations
of existing replication-based protocols (Epidemic, Spray and Wait, Spray and Focus), proba-
bilistic (PRoPHET), and social-based protocols (Bubble Rap, SCORP, dLife, dLifeComm) in
the OppNet. Besides, node density and message time-to-live (TTL) have important roles in
message forwarding in the OppNet. With the increase of node density, the network size is also
extended, and the increase of TTL means the message will be active for a longer time. These
may affect the performance of the OppNet. In this research, we have investigated for a more
energy-efficient protocol among the aforementioned protocols. A brief performance analysis
of these protocols is also addressed in this study. The entire experimental evaluation has been
performed from the impact of both node density and TTL after considering four performance
criteria: average remaining energy, delivery ratio, average latency, and overhead ratio.
The structure of this chapter is as follows: a brief description of various OppNet routing
protocols from various perspectives are provided in Section 11.2. Section 11.3 provides a
general explanation of the simulation tool and settings. Performance metrics that are essen-
tial to our study are defined in Section 11.4. A comparative discussion on simulation out-
comes is given in Section 11.5, and finally, Section 11.6 gives the conclusion of our study.

11.2 ROUTING PROTOCOLS FOR OppNet

The main objective of a routing protocol for OppNet is to deliver the message to the des-
tination node successfully from a message carrier node. This forwarding operation should
act to increase the message delivery rate and to reduce the latency, overhead, and number of
message copies [20]. Furthermore, a routing protocol should be energy-efficient since nodes’
energy is restricted. So, for better networking performance, less energy should be consumed
by each node [19]. Several routing protocols have been developed for the OppNet. Stochastic-
natured routing protocols are the main priority for data forwarding in most of the OppNet
scenarios. Here, determining the future position of nodes is difficult, and the characteristics
of the network are unpredictable [20]. However, routing protocols have been designed from
various networking aspects to meet the requirements of such a challenging environment as
the OppNet. In this chapter, we have considered replication-based (i.e., Epidemic, Spray and
Wait, and Spray and Focus), probabilistic (i.e., PRoPHET), and social-based protocols (i.e.,
Bubble Rap, SCORP, dLife, and dLifeComm) for our studies, and a brief discussion of these
protocols is provided in this section.

11.2.1 Replication-Based Protocols
Message replication-based protocols are the first-generation protocols for OppNet. The
replication-based protocol can be referred to as a flooding-based protocol. In this case, the
source node sends multiple copies of a message to the encountered nodes. Thus, flooding of
the same message can occur in the network. The replication-based protocols have two major
categories [21]:

• Uncontrolled replication: When a source node sends an unlimited replica of a message

to all possible relays, this situation is referred to as uncontrolled replication. Epidemic
is a replication-based protocol in which an uncontrolled replication technique is
exploited.
210  Opportunistic Networks

• Controlled replication: In this case, a limited number of message replicas have been sent
to a limited number of relays. Spray and Wait, and Spray and Focus protocols adopt
this strategy to forward messages.

11.2.1.1 Epidemic
Epidemic is the first representative of the replication-based OppNet protocol. The major
issue behind this forwarding strategy is copying messages. This strategy deploys blind repli-
cation of messages. The source node repeatedly copies its message and sends all the message
copies to the encountered nodes who have not received any copy of the message yet. In this
manner, the message is distributed into the full network, and finally, the target node will get
the message. In Epidemic strategy, successful message delivery would be possible without
regarding some issues such as congestion of network, buffer capacity, delivery delay, etc. Such
unawareness behavior about the resources of Epidemic makes it a weak candidate for data
routing in OppNet [21, 22].

11.2.1.2 Spray and Wait

Spray and Wait comes from the controlled replication genre. In this strategy, the blind replica-
tion of Epidemic has been replaced by controlled replication. So, the utilization of resources is a
minimum here. In Spray and Wait, L number of message replicas are spread into the whole net-
work, where L denotes the maximum number of copies of a message that are allowed to route
from the source. The value of L can be fixed based on the number of nodes and the average
delay that is required to deliver a message [18]. This protocol is composed of two phases [5]:

• Phase 1: This phase is known as the “Spray” phase, where the source node will forward
L copies of a message towards L distinct encountered nodes with the source.
• Phase 2: If the destination node is not found in the “Spray” phase, then the message
accepting L distinct nodes in the previous phase will wait until they can encounter the
destination node to forward the message copy directly. This phase is known as “Wait”.

Basically, the Spray and Wait forwarding protocol has two modes: Binary and Vanilla. The
major differences between these two modes exist in the basic strategy adopted in the “Spray”
phase to spray L replica of a message. In “Binary” mode, the message carrying source node
begins with L copy of a message. The source node simply forwards L/2 copy of the message
to that node who first encountered the source. Then each encountered message receiving
node delivers half of the message copies carried by it toward those nodes they may meet in
the future and those have no message copy. Finally, a node switches to the “Wait” phase while
it delivers all of the message copies except one. In the “Wait” phase, it will wait for the direct
opportunity to encounter with destination node for successful message transmission. In con-
trast, another mode named “Vanilla” where the source node would deliver a single copy of
the message toward the L distinct relay nodes which encounter the source node after the gen-
eration of that message in the source. However, the “Binary” mode has the advantage over
the “Vanilla” mode in Spray and Wait, and this advantage is that the message is disseminated
much faster than in “Vanilla” mode [5]. In this study, only the “Binary” mode is considered
for the Spray and Wait routing protocol.

11.2.1.3 Spray and Focus

Spray and Focus is the modified version of the Spray and Wait protocol. This forwarding
technique has set up an upgraded single-copy routing strategy. Within this technique, once
 Evaluation of Energy Efficiency  211

a relay node receives a single replica of a message, it would retransmit that copy instead of
direct delivery. This routing strategy also consists of two phases: “Spray” and “Focus”. Here,
the first phase is similar to the Spray and Wait protocol. But the dissimilarity between the two
takes place in the “Focus” phase. Unlike the “Wait” phase in Spray and Wait, relay nodes are
not waiting for the destined node to encounter. They simply forward a single message copy
toward a more consistent relay node. This forwarding is performed according to a utility-
driven model. For a distinct message, if there is any relay with a token of single transmission,
it would be turned into the “Focus” phase. So, a particular relay node may be able to receive
a replica of the message in the “Focus” phase, instead of the destination node receiving it
through direct transmission performed in the “Wait” phase of Spray and Wait. Finally, the
functionalities of Spray and Focus protocols can be explained as follows [6]:

• Message summary vector: A vector is maintained by each node which contains the IDs
of all messages stored by it and also those for which it acts as a relay. When two nodes
encounter each other, they will swap their vectors. Then, they check the common mes-
sage between them, and this message has a lifetime. The message would be discarded
if this lifetime ends, and the entry of such discarded message would be erased in the
message summary vector.
• Last encounter timers: A timer τi (j) for every other node j in the network is maintained
by each node denoted by i. Let tm (d) denote the required time in which a node has to
move a distance d under a given mobility model, m. If a node A encounters a node B at
distance dAB, then the following condition must be satisfied:

i  j  j  i   0;

j  B : B  j  A  j  tm  d AB  ,

Where, A  j  B  j  tm  d AB  .
• Spray and Focus forwarding: At the source node, while a message is created, L “for-
warding tokens” are also created in the source. If a node, either source or relay, carries
a message copy, it would forward the message based on either one of two conditions:
i. n > 1 forwarding tokens, in which binary spraying would be performed; or,
ii. n = 1 forwarding token, which would perform utility-based forwarding according
to the last encounter timers used as the utility function; here, n denotes the number
of forwarding tokens.

11.2.2 Probabilistic Protocols
To address the efficient message forwarding problem in OppNet, researchers have intro-
duced probabilistic concepts in the forwarding decision. Although a dedicated and fully
connected end-to-end link is not guaranteed in OppNet, in reality, there are many cases
where the probability of communication is nevertheless expected. So, this point is considered
as the background behind the implementation of the probabilistic idea in data routing for
the OppNet environment to improve the success rate of message delivery. The Probabilistic
Routing Protocol using History of Encounters and Transitivity (PRoPHET) is an instance of
this protocol class [7].

11.2.2.1 PRoPHET
The Probabilistic Routing Protocol using History of Encounters and Transitivity (PRoPHET)
algorithm ensures the proper utilization in OppNet. To make sure that the perfect message
212  Opportunistic Networks

delivery is performed on the proper destined node, a set of probabilities is conserved within
this routing algorithm of PRoPHET. Unlike the blind replication property of the uncontrolled
replication-based protocol, the likelihood of the nodes’ real-world confrontation is adopted
by this algorithm for efficient delivery of a message. After the encounter of other nodes with
the message carrier source node, the message would be forwarded to the destined node fol-
lowing the higher probability of message transmission. The message is received first by that
particular node which has a higher probability of transmission [7, 18].

11.2.3 Social-Based Protocols
In many OppNet environments, a large number of mobile nodes may be allocated in a scat-
tering manner within a small and dense area. Thus, the topology is fully heterogeneous and
forwarding protocols of the previous generation may not be suitable to route data effectively
in such an application. Social-based protocols have presented a new generation of forward-
ing protocols to cope with this challenge by exploiting some social metrics such as commu-
nity, centrality, etc., in forwarding decisions. The social attitudes and characteristics of node
groups and the reflections of their inter-relationship are also recognized to select the latest
eligible relay node. Thus, nodes interact in such divergent ways with each other where some
nodes are used to more frequent encounters in a social-based network. However, nodes’
social behaviors and attributes are the long-term properties, and this seems to be less volatile
in comparison with nodes’ mobility. Bubble Rap, SCORP, dLife, and dLifeComm protocols
belong to this protocol type [23, 24].

11.2.3.1 Bubble Rap
This forwarding technique involves both communities as well as centrality. A collection of
nodes can form a community. Within it, there are a few nodes that have greater acceptance
than others on behalf of social interactions. Thus, these nodes may be regarded as having
higher centrality. A particular node with the higher centrality may be selected as the eligible
next relay by a message-containing source node to forward with a view to deliver towards the
targeted final node. Nevertheless, if a Bubble Rap-implemented community in the OppNet
scenario becomes extended by increasing the density of nodes with lower centrality, then the
success rate of message delivery is assumed to be a minimum. So, the Bubble Rap forwarding
technique is not applicable for pervasive OppNet scenarios, because to estimate the central-
ity for nodes, overall information relevant to the total network topology is needed. For this
reason, the computation of centrality is not practicable for a spreading distributed network
where a large number of nodes are employed [10, 25].

11.2.3.2 SCORP
The Social-Aware Content-Based Opportunistic Routing Protocol (SCORP) is another social-
aware forwarding protocol where the interest of message content is the major concerning
issue. In this strategy, knowledge about message content is required by each node including
the type of message content, interested hosts, etc. A routing decision in SCORP is taken to
identify the next relay depending on the content interest as well as nodes’ social proximity.
Here, a message owner source node forwards the message to that distinct encountered node
which has a similar interest in the content of the message as the source node. Alternatively,
the node which has a higher degree of relations with the message-containing node will have
priority to receive the message first [8, 18].
 Evaluation of Energy Efficiency  213

11.2.3.3 dLife
dLife is a social-based forwarding approach which is based on the regular activities of users’
everyday lives. To understand users’ daily dynamic activity, a weighted social graph is required
to be prepared. In this strategy, the source node routes a single message copy to that encoun-
tered node which has a strong relationship with the target node. So, there exists a prob-
ability that in the future it will encounter the target node, and finally, the message reaches
the destined node. Furthermore, if this relationship between the source and target node is
unexplored, then the message would forward in accordance with the importance of the node.
The source node will forward the message to the most important node, and this importance
is computed by source [9, 18, 25].

11.2.3.4 dLifeComm
A community variant of dLife protocol is dLifeComm. The main concept of forwarding data
in this strategy is quite similar to the dLife approach. Here, the routing decision is made on
the basis of social strength within a community, and it is calculated from the source and
encountered nodes towards the destined node. Additionally, to forward the message outside
a community, node importance is considered, and this protocol is focused mainly on the
community rather than the centrality. The K-Clique algorithm is adopted in this forwarding
strategy to detect such communities [9].

11.3 SIMULATION ENVIRONMENTS

This section provides a brief discussion about the environment needed to simulate the inves-
tigated forwarding protocols for OppNet.

11.3.1 Simulation Tool
The Opportunistic Network Environment (ONE) simulator is adopted as a simulation tool
where all protocols are simulated. ONE is a Java-based simulator that is specially designed
for the opportunistic environment. Various lucrative features of ONE simulator, such as visu-
alizing node mobility, endorsing new maps and mobility models, and generating reports after
simulation, make it an ideal tool for simulating OppNet scenarios [26]. Figure 11.1 provides
a view of the ONE simulator. In this chapter, selected OppNet forwarding protocols from
various genres have been simulated with a similar environment in the ONE simulator to
investigate the energy efficiency along with their relevant performance comparison.

11.3.2 Simulation Scenario Settings

The entire experimental area is assumed as 10 km × 10 km on the Helsinki city map, which
is the default in ONE simulator. The total duration of the simulation is considered to be
6 hours. Three groups of nodes are used, such as pedestrian, tram, and car. The average
velocity (minimum and maximum) of pedestrian, tram, and car nodes in each group is set to
0.8–1.4, 2.7–13.9, and 7–10 m/s respectively. Here, nodes are communicating through the
Bluetooth interface, and the maximum transmission range of each node is 100 m. Nodes’
movement is performed according to the shortest path map-based movement. Eight differ-
ent OppNet routing protocols have experimented into this study such as Epidemic, Spray
and Wait, Spray and Focus, PRoPHET, Bubble Rap, SCORP, dLife, and dLifeComm. These
214  Opportunistic Networks

Figure 11.1  A view of the ONE simulator.

protocols are simulated in the ONE simulator by varying the node density (50, 80, 110, 140,
170, and 200) while TTL is kept fixed at 360 min (6 hours). Again, simulations are performed
by varying time-to-live (TTL) from 120 to 360 min while keeping the node density per group
fixed (50 nodes in each group). Table 11.1 exhibits the lists of necessary parameters that are
required to perform simulations. Here, a maximum of 10 allowable copies is considered for
Spray and Wait as well as Spray and Focus, and the number of seconds in the time unit is
fixed at 30 for PRoPHET.

Table 11.1  Necessary settings for simulation arrangements.

Parameter Value

Simulation duration 6h
Node density 50, 80, 110, 140, 170, 200
Interface type Bluetooth
Transmission rate 250 kbps
Transmission range 100 m
Routing Protocols Epidemic, Spray and Wait, Spray and Focus, PRoPHET, Bubble Rap, SCORP, dLife,
dLifeComm
Buffer capacity (MB) 5
Message TTL (in minutes) 120, 180, 240, 300, 360
Message size 500 KB to 1 MB
Mobility model Shortest path map-based movement
Simulation area size 10 km × 10 km
 Evaluation of Energy Efficiency  215

Moreover, to examine the energy efficiency of experimented protocols, an energy module

[27] is also implemented in the ONE simulator. Table 11.2 provides the necessary energy set-
tings of nodes. Here, the initial battery energy of each node is given 5000 J before simulation.
The energy expenditure which is spent per second on each scan for the nearest device discov-
ery is known as “Scan energy”. The amount of energy that is consumed per second in the time
of message transmission is known as “Transmit energy”. “Receive energy” can be denoted as
the energy quantity expensed per second during the message reception in each receiving node.
The time interval within which a node gets recharged is known as the “Interval of energy
recharge”. In this study, this time interval is regarded as 28,000 s, or 7.78 hours, which is
greater than the entire simulation time (6 hours). This makes sure that the energy-constrained
mobile nodes would not be able to get recharged during the overall simulation since they
have been charged once before the start of the simulation.

11.4 PERFORMANCE METRICS

After performing simulations, the outcomes of simulated protocols are evaluated on behalf
of four performance metrics: average remaining energy of each node, delivery ratio, average
latency, and overhead ratio. This section briefly describes these metrics.

11.4.1 Average Remaining Energy

The average quantity of energy that remains in each node after fully completing the simula-
tion is known as the average remaining energy. The value of this metric is expected as higher.
A protocol that is implemented in an OppNet scenario can be considered as energy-efficient
if the nodes of that network have higher remaining energy.

11.4.2 Delivery Ratio
The delivery ratio can be defined as the ratio of the message amount that successfully reached
the destined node to the amount of message produced by the source node. The probability of
successful message delivery is reflected in an OppNet by this metric. So, a higher success rate
of message delivery depends on a higher delivery ratio.

11.4.3 Average Latency
This metric is defined as the average delay time required between a single message generation
in the source node and its successful reception in the destination. For efficient OppNet opera-
tion, the value of this metric needs to be a minimum.

Table 11.2  Parameters related to energy of node.

Parameter Settings

Capacity of battery 5000 J

Scan energy 0.92 mW/s
Transmit energy 0.08 mW/s
Receive energy 0.08 mW/s
Interval of energy recharge 28,000 s
216  Opportunistic Networks

11.4.4 Overhead Ratio
The overhead ratio measures how many redundant packets are relayed to deliver a single
packet to the destination. Thus, it reflects the cost of transmission in OppNet, and the value
of this metric is required to be a minimum.

11.5 RESULTS AND DISCUSSION

A comparative analysis of simulated OppNet protocols is given in this section regarding

energy efficiency and performance evaluations. The discussions that follow are performed
through the impact of both node density and TTL.

11.5.1 Impact of Node Density

Figure 11.2 exhibits the average energy of each node remaining after simulation with the
increase of node density in each group. The increase of node density means the scanning,
transmitting, and receiving operations are also increased, so energy would be consumed more.
Here, SCORP and dLife both have higher remaining nodes’ energy than other experimented
OppNet protocols. The other two social-based protocols – Bubble Rap and dLifeComm
protocols – have the average node energy (less than Spray and Wait, and less than Spray and
Focus, but higher than Epidemic). The improved forwarding strategies of SCORP and dLife
make them more energy-efficient, as both forwarding strategies consider interests of message
content and daily behaviors of users, respectively. SCORP has the highest energy efficiency.
So in SCORP, the least amount of energy is consumed for successful message delivery. On the
other hand, Epidemic has the minimum energy efficiency than others because of the unaware-
ness of resource (energy) during message forwarding. Thus, more energy is consumed by
nodes for unnecessary as well as countless message replication and delivery to the next relays.
Moreover, Spray and Wait, Spray and Focus, and dLifeComm protocols experienced less
remaining node energy than SCORP and dLife. But these protocols have a higher value of
remaining node energy than other simulated OppNet protocols (both PRoPHET and Bubble

Figure 11.2  Average remaining energy vs. node density in each group.
 Evaluation of Energy Efficiency  217

Rap, respectively) with the variation of node density. Among these three protocols, the Spray
and Wait protocol has much better performance for average remaining node energy, since a
limited number of message copies is forwarded to the relay node. Additionally, Spray and
Focus exhibits less remaining energy of each node than Spray and Wait, as more energy is
invested by each node to further encounter with destination node. So, it can be decided that
Spray and Wait, Spray and Focus, dLifeComm, PRoPHET, and Bubble Rap show the medium
performance in this scenario in terms of average remaining energy, respectively, while the
node density is increased.
From Figure 11.3, it can be clarified that dLife, Spray and Focus, and Spray and Wait have
the highest delivery ratio with increasing node density in each group. But in comparison, both
dLife and Spray and Focus have higher delivery ratios, and Spray and Wait experienced the
second-highest performance among all simulated protocols. This is possible due to the selec-
tion of the next relay for dLife depending on either the relations with destination or node
importance. So, the success ratio of message delivery to the destination node is higher. Like
dLife, Spray and Focus has also the highest delivery ratio, because there are fewer message
replicas here, and the relay nodes who receive message replicas would be intended for the
final delivery of messages toward the destination node. So, the delivery ratio is the highest in
this case. Again, the limited message replication process and selecting the proper relay nodes
to forward messages in Spray and Wait protocol make delivery ratio higher in this case. But
as compared to the best performers (dLife, Spray and Focus), the relay nodes who received
message in the “Spray” stage would wait for the destination node for the final encounter.
These activities of intermediate relay nodes may reduce the delivery ratio of Spray and Wait
than Spray and Focus as well as dLife. The Spray and Wait protocol exhibits the second-best
performance in terms of delivery ratio. Oppositely, Epidemic has the lowest performance with
increasing node density because of its implementation of the flooding technique. So, many
redundant message copies are forwarded from hop to hop in Epidemic, which results in the
reduction of message delivery probability. The delivery ratio of another social-based proto-
col, dLifeComm, fluctuates with the increase of node density. The delivery ratio of SCORP
remains constant with the extension of nodes per group. So, we can consider SCORP as a
medium performer for delivery ratio. PRoPHET and Bubble Rap both have a minimum deliv-
ery ratio as compared to other protocols except the worst-performing protocol, Epidemic.

Figure 11.3  Delivery ratio vs. node density in each group.

218  Opportunistic Networks

The variation of average latency with varying node density is demonstrated in Figure 11.4.
From this figure, it is realized that dLifeComm has the highest delivery delay if the network
size is expanding. Because, in a community, it is possible that message would be delivered
with delay to the known node. Moreover, the dLife protocol has the second-highest latency,
since there may be a tendency of the nodes to deliver the message with delay towards those
who are closely related to the destination node. Here, both dLifeComm and dLife protocols
have comparatively higher latency than other simulated OppNet protocols. Spray and Focus
has the lowest latency due to its sophisticated forwarding strategy, which makes it the best-
performing protocol here. Simulated OppNet protocols like Bubble Rap, PRoPHET, Spray
and Wait, Epidemic, and SCORP demonstrated medium performance in this case. The aver-
age delay time which is necessary to deliver a message in these protocols is not more than 30
minutes. Message forwarding in these protocols would take medium delay time as compared
to dLifeComm, dLife (maximum delay time), and Spray and Focus (minimum delay time).
The overhead ratio is plotted with the variation of node density per group, as shown
in Figure 11.5. Here, it is observed that Epidemic has the highest overhead ratio since the

Figure 11.4  Average latency vs. node density in each group.

Figure 11.5  Overhead ratio vs. node density in each group.

 Evaluation of Energy Efficiency  219

unbound flooding of messages is deployed there. As a significant number of message copies

have been wasted for the successful delivery of messages to the destinations, the overhead
ratio is increased gradually with the increase of network size. Thus, Epidemic has the worst
performance for overhead ratio. On the other hand, SCORP, Spray and Wait, Spray and
Focus, dLifeComm, and dLife protocols (listed in order of performance) have the minimum
overhead ratio. So, these protocols have the best performance for overhead ratio. Along
with that, both PRoPHET and Bubble Rap have medium performance as compared to the
best- and worst-performing protocols. Between these two protocols, the Bubble Rap protocol
exhibits a higher overhead ratio than PRoPHET. Bubble Rap becomes ineffective when the
size of the network is larger, due to the increase of node density.

11.5.2 Impact of TTL
In Figure 11.6, the average remaining energy is plotted with varying TTL (from 120 to
360 min). The increase in the value of message TTL means the longevity of a message would
also increase so that the message can be active for a longer time. More energy would be
required to deliver more active messages. As with the results of Figure 11.2, both SCORP and
dLife have higher remaining energy than the others. Successful message delivery is possible
for SCORP and dLife due to the effective forwarding techniques whether the message lifetime
is increased. So, SCORP and dLife are the best performers on behalf of average remaining
energy. Furthermore, SCORP exhibits better performance than dLife. In contrast, Epidemic
exhibits the worst performance in this case due to resource unaware replications. Spray and
Wait, Spray and Focus, and dLifeComm (listed in order of performance) are the medium
performers in this case. PRoPHET and Bubble Rap exhibit slightly better performance than
the worst performer, Epidemic, and between these two protocols, PRoPHET performs quite
better than Bubble Rap.
The impact of TTL on the delivery ratio is visualized in Figure 11.7. In this figure, Spray
and Focus and dLife have the highest ratio of message delivery, so we can consider both
these forwarding protocols as the best performers among other examined OppNet protocols.
Spray and Wait has a slightly lower value of delivery ratio than Spray and Focus and dLife,
since the Spray and Wait indicating line is positioned below those of the other two protocols.
It is also realized from this figure that the delivery ratio of dLifeComm is growing with the
increase of TTL and when TTL is 240 min, the delivery ratio of dLifeComm becomes higher

Figure 11.6  Average remaining energy vs. TTL.

220  Opportunistic Networks

Figure 11.7  Delivery ratio vs. TTL.

as do Spray and Focus and dLife, since in a community the rate of successful message deliv-
ery would increase if message TTL is increased. Consequently, similar to the results of the
node density impact, the delivery ratio is the lowest for Epidemic among the others. If the
redundant message copies have a larger lifetime, it would not be effective for the successful
transmission of those message copies.
Figure 11.8 represents the same result which is provided by Figure 11.4 for the best-
performing protocol. For both cases, Spray and Focus experienced the lowest latency to
deliver a successful message towards the destination. So, Spray and Focus performs the best
among all simulated protocols. A limited number of message copies are successfully delivered
to the destination with minimum delay. If the message lifetime is increased here, the delay
time between the generation of a message in the source and its successful reception at the
destination remains at a minimum. On the other hand, dLife and dLifeComm have a higher
latency than others. But between these two protocols, dLife has quite higher latency than
dLifeComm, So, it can be considered as the worst-performing protocol of all. Remaining
protocols such as Bubble Rap, PRoPHET, SCORP, Epidemic, and Spray and Wait exhibit the
medium performance, respectively.
From Figure 11.9, it is derived that Epidemic experienced the highest value for overhead
ratio. This finding is similar to Figure 11.5. The impact of both increasing node density and

Figure 11.8  Average latency vs. TTL.

 Evaluation of Energy Efficiency  221

Figure 11.9  Overhead ratio vs. TTL.

TTL is similar for Epidemic. This protocol has the maximum overhead to deliver a single
message to the destination. Many unnecessary message copies are generated by the source
node in Epidemic. So, the ratio of overhead is increased and the network becomes congested
for Epidemic protocol with the rise of message TTL. In opposition, three protocols – SCORP,
Spray and Wait, and Spray and Focus – have minimum overhead than other OppNet proto-
cols. So, these protocols can be recognized as the best-performing protocols for this metric,
and their values of the overhead ratio are almost zero. So, very few messages would be
wasted during message transmission for these protocols. This condition would not be vio-
lated if the message were to live for an additional time. Moreover, dLife and dLifeComm
jointly exhibited a slightly higher value of overhead ratio than the best-performing protocols,
and their indicating graph line is positioned slightly higher than the three best-performing
protocols. Since within these two protocols, fewer messages would be wasted for a successful
message delivery whenever the message lifetime grew. But this amount is slightly higher than
the best-performing protocols SCORP, Spray and Wait, and Spray and Focus.

11.6 CONCLUSIONS

Within this chapter, we have evaluated the energy efficiency as well as the performance analy-
sis of eight different OppNet forwarding protocols: replication-based (Epidemic, Spray and
Wait, Spray and Focus), probabilistic (PRoPHET), and social-based (Bubble Rap, SCORP,
dLife, dLifeComm). These evaluations were performed on the basis of varying both node
density and TTL impact upon four performance metrics: average remaining energy, delivery
ratio, average latency, and overhead ratio. Simulations of the mentioned OppNet protocols
have been done on the ONE simulator. After analyzing the outcomes, we can conclude that
with increasing node density and TTL, SCORP is the most energy-efficient protocol, while
Epidemic is the least energy-efficient. Furthermore, dLife and Spray and Focus have the maxi-
mum delivery ratio. Moreover, Spray and Focus is the best-performing protocol for average
latency. dLife (for TTL variation) and dLifeComm (for node density variation) have the least
performance for average latency. For overhead ratio, SCORP, Spray and Focus, and Spray
and Wait are the best-performing strategies, whereas Epidemic offers the worst performance
for delivery probability and overhead ratio, respectively.
222  Opportunistic Networks

ACKNOWLEDGMENTS

This work has been supported by National Science and Technology (NST) MPhil fellowship,
granted by Ministry of Science and Technology, Government of the People’s Republic of
Bangladesh.

REFERENCES

[1] Pelusi, L., Passarella, A., and Conti, M. (2006). Opportunistic networking: data forwarding in
disconnected mobile ad hoc networks. IEEE Communications Magazine, 44(11), pp. 134–141.
[2] Huang, C. M., Lan, K. C., and Tsai, C. Z. (2008). A survey of opportunistic networks. The 22nd
International Conference on Advanced Information Networking and Applications-Workshops,
pp. 1672–1677.
[3] Trifunovic, S., Kouyoumdjieva, S. T., Distl, B., Pajevic, L., Karlsson, G., and Plattner, B. (2017). A
decade of research in opportunistic networks: challenges, relevance, and future directions. IEEE
Communications Magazine, 55(1), pp. 168–173.
[4] Vahdat, A., and Becker, D. (2000). Epidemic routing for partially connected ad hoc networks.
Department of Computer Science, Duke University, Technical Report, CS-2000-06.
[5] Spyropoulos, T., Psounis, K., and Raghavendra, C. S. (2005). Spray and Wait: an efficient routing
scheme for intermittently connected mobile networks. Proceedings of the 2005 ACM SIGCOMM
Workshop on Delay-Tolerant Networking, pp. 252–259.
[6] Spyropoulos, T., Psounis, K., and Raghavendra, C. S. (2007). Spray and Focus: efficient mobility-
assisted routing for heterogeneous and correlated mobility. Fifth Annual IEEE International
Conference on Pervasive Computing and Communications Workshops (PerComW'07), pp.
79–85.
[7] Lindgren, A., Doria, A., and Schelén, O. (2003). Probabilistic routing in intermittently con-
nected networks. ACM SIGMOBILE Mobile Computing and Communications Review, 7(3),
pp. 19–20.
[8] Moreira, W., Mendes, P., and Sargento, S. (2014). Social-aware opportunistic routing protocol
based on user’s interactions and interests. International Conference on Ad Hoc Networks, pp.
100–115. Springer, Cham.
[9] Moreira, W., Mendes, P., and Sargento, S. (2012). Opportunistic routing based on daily routines.
2012 IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks
(WoWMoM), pp. 1–6.
[10] Hui, P., Crowcroft, J., and Yoneki, E. (2010). Bubble Rap: social-based forwarding in delay-
tolerant networks. IEEE Transactions on Mobile Computing, 10(11), pp. 1576–1589.
[11] Jagtap, P., and Kulkarni, L. (2019). Social energy-based techniques in delay-tolerant network.
Emerging Technologies in Data Mining and Information Security, pp. 531–538. Springer,
Singapore.
[12] Vardalis, D., and Tsaoussidis, V. (2014). Exploiting the potential of DTN for energy-efficient
internetworking. Journal of Systems and Software, 90, pp. 91–103.
[13] Socievole, A., and Marano, S. (2012). Evaluating the impact of energy consumption on routing
performance in delay tolerant networks. 8th International Wireless Communications and Mobile
Computing Conference (IWCMC), pp. 481–486.
[14] Cabacas, R. A., Nakamura, H., and Ra, I. H. (2014). Energy consumption analysis of delay
tolerant network routing protocols. International Journal of Software Engineering and Its
Applications, 8(2), pp. 1–10.
[15] Kaviani, M., Kusy, B., Jurdak, R., Bergmann, N., and Liu, V. (2016). Energy-aware forward-
ing strategies for delay tolerant network routing protocols. Journal of Sensor and Actuator
Networks, 5(4), p. 18.
[16] Bista, B. B., and Rawat, D. B. (2016). Energy consumption and performance of delay tolerant
network routing protocols under different mobility models. 7th International Conference on
Intelligent Systems, Modeling and Simulation (ISMS), pp. 325–330.
 Evaluation of Energy Efficiency  223

[17] Spaho, E. (2019). Energy consumption analysis of different routing protocols in a delay tolerant
network. Journal of Ambient Intelligence and Humanized Computing, 11, pp. 3833–3839.
[18] Khan, M. K. M., and Rahim, M. S. (2018). Performance analysis of social-aware routing proto-
cols in delay tolerant networks. 2018 International Conference on Computer, Communication,
Chemical, Material and Electronic Engineering (IC4ME2), pp. 1–4.
[19] Khan, M. K. M., Roy, S. C., Rahim, M. S., and Islam, A. Z. M. T. (2020). On the energy effi-
ciency and performance of delay-tolerant routing protocols. Lecture Notes of the Institute for
Computer Sciences, Social Informatics and Telecommunications Engineering, 325, pp. 553–565.
Springer, Cham.
[20] Mota, V. F., Cunha, F. D., Macedo, D. F., Nogueira, J. M., and Loureiro, A. A. (2014). Protocols,
mobility models and tools in opportunistic networks: a survey. Computer Communications, 48,
pp. 5–19.
[21] Aloui, E. A. A., Said, A., and Moha, H. (2015). The performance of DTN routing protocols: a
comparative study. WSEAS Transactions on Communications, 14, pp. 121–130.
[22] Ababou, M., and Bellafkih, M. (2018). Energy efficient routing protocol for delay tolerant
network based on fuzzy logic and ant colony. International Journal of Intelligent Systems and
Applications, 10(1), pp. 69–77.
[23] Qirtas, M. M., Faheem, Y., and Rehmani, M. H. (2020). A cooperative mobile throwbox-
based routing protocol for social-aware delay tolerant networks. Wireless Networks, 26, pp.
3997–4009.
[24] Zhang, X., Huang, P., Guo, L., and Fang, Y. (2019). Social-aware energy-efficient data offloading
with strong stability. IEEE/ACM Transactions on Networking, 27(4), pp. 1515–1528.
[25] Haq, A., and Faheem, Y. (2020). A peer-to-peer communication based content distribution pro-
tocol for incentive-aware delay tolerant networks. Wireless Networks, 26(1), pp. 583–601.
[26] Keränen, A., Ott, J., and Kärkkäinen, T. (2009) The one simulator for DTN protocol evaluation.
2nd International Conference on Simulation Tools and Techniques (SIMUTools ‘09), pp. 1–10.
[27] Opportunistic Network Environment (ONE) simulator project page. https://www.netlab.tkk.fi/
tutkimus/dtn/theone. Accessed 7 November 2020.
Chapter 12

Mobility Models in Opportunistic Networks

Jagdeep Singh
Sant Longowal Institute of Engineering and Technology, Longowal, India

Sanjay Kumar Dhurandher

Netaji Subhas University of Technology, New Delhi, India

Vinesh Kumar
University of Delhi, New Delhi, India

CONTENTS

12.1 Introduction............................................................................................................. 226

12.2 Trace-Based Model.................................................................................................. 227
12.2.1 Trace-Based Analysis.................................................................................. 227
12.3 Stochastic Mobility Model....................................................................................... 228
12.3.1 Random-Based Mobility Models................................................................ 228
12.3.1.1  Random Walk Model.................................................................. 228
12.3.1.2  Random Direction Model........................................................... 228
12.3.1.3  Random Waypoint Model........................................................... 229
12.4 Synthetic Models..................................................................................................... 229
12.4.1 Temporal Dependency-Based Mobility Models........................................... 229
12.4.2 Spatial Dependency-Based Mobility Models............................................... 230
12.5 Geographical Restriction-Based Mobility Model..................................................... 230
12.5.1 Pathway Mobility Model............................................................................ 230
12.5.2 Obstacle Mobility Model............................................................................ 230
12.5.3 Manhattan Mobility Model........................................................................ 231
12.5.4 City Segment Model................................................................................... 231
12.6 Map-Based Mobility Models................................................................................... 231
12.6.1 Route-Based Map Mobility Model............................................................. 231
12.6.2 Rush Hour (Human) Traffic Model............................................................ 231
12.6.3 Working Day Movement Model................................................................. 232
12.6.4 Shortest Path Map-Based Movement Model............................................... 233
12.6.5 Social Network-Based Mobility Models..................................................... 233
12.6.6 Community-Based Mobility Models........................................................... 234
12.6.7 Social Network Models.............................................................................. 234
12.7 Simulation Analysis .................................................................................................234
12.7.1 Testing Tool................................................................................................ 236
12.7.2 Simulation Results...................................................................................... 236
12.8 Conclusion............................................................................................................... 240
References........................................................................................................................ 241

225
226  Opportunistic Networks

12.1 INTRODUCTION

Due to the randomly disconnected topology, mobility plays a significant role in opportunistic
networks (OppNets), as the mobility characteristics are used to formulate the mobility mod-
els in a realistic scenario. From the analysis and simulation, it is observed that the mobility
models based on realistic topology have an important place in OppNets. The unrealistic sce-
nario paves to unrealistic behavior or results when the mobility models based on unrealistic
topology are used for a real-world scenario. Therefore, these unrealistic mobility models do
not perform well in the real world. The evolution of a system in OppNets basically utilizes
traffic patterns and mobility models [1, 2].
Mobility characteristics include the speed, predictability of movement patterns, and uni-
formity of mobile nodes in communication networking. Thus, the characteristics and features
are the basis for the classification of mobility models. There exist three types of mobility
models, namely trace-based models, synthetic models, and stochastic models. In conventional
MANETs, networking protocols were generally evaluated using stochastic movement mod-
els as the Random Walk Mobility model (RWM), the Random Waypoint mobility model
(RWP), the Random Direction Mobility model (RDM), etc. [3]. However, users’ mobility
is dependent upon the users’ movement behavior, their social and personal characteristics,
and environmental factors, because in reality users’ movement is rarely random; thus, the
conventional stochastic random-based mobility models fail to estimate the accuracy of the
networking protocols in OppNets. In the literature, various studies demonstrate that users’
social and personal behavior has a significant correlation with the users’ movement patterns
[4]. For instance, in recent studies and experimental analysis, it has been proven that users
frequently visit a few places where they stay most of their time and with which they have
strong social relationships. Furthermore, they rarely travel long distances and very often
travel over short distances. Hence, based on the mobility characteristics discussed in the fol-
lowing section, an appropriate selection of mobility traces and models is necessary to emu-
late the movement behavior of humans and vehicles. Mobility models are categorized into
four categories: a trace-based model of mobility, a stochastic model of mobility, a map-based
model of mobility, and a synthetic model of mobility. Figure 12.1 represents the classifica-
tions for the mobility models.

Figure 12.1  Mobility models classification.

 Mobility Models in Opportunistic Networks  227

The remainder of the chapter is discussed as follows. Section 12.2 covers the trace-based
model. A detailed overview of models of stochastic mobility is given in Section 12.3. In
Section 12.4, synthetic models are discussed. The geographical restriction dependent mobil-
ity model is introduced in Section 12.5. Section 12.6 focuses on the map-based mobility
model. The findings and interpretation of the simulation are discussed in Section 12.7. Lastly,
Section 12.8 concludes the chapter.

12.2 TRACE-BASED MODEL

12.2.1 Trace-Based Analysis
Trace-based models [5] can have an insight into actual mobility patterns by tracing
mobile hosts in real-world scenarios. There are three methods for acquiring traces. The
first method is to trace the location of devices by monitoring them with a particular
tool. Currently, Global Positioning System (GPS) is used for monitoring the localization
system. The second method is to use the communication system for monitoring com-
munication devices. The accuracy of monitoring the communications is based upon the
density of the access point to which communication devices communicate. This method
may not be precise, but it can be used to validate the mobility models. The third method
is to acquire contact traces using Bluetooth or WLAN by monitoring the contacts among
mobile devices (Table 12.1).

Table 12.1  List of abbreviations used in the chapter.

Acronym Description

GPS Global Positioning System

RWM Random Walk Model
RDM Random Direction Model
RWP Random Waypoint Model
SPMBM Shortest Path Map-Based Movement Mobility Model
ONE Opportunistic Network Environment
WLAN Wireless Local Area Network
WDM Working Day Movement Model
CMM Community-Based Mobility Model
POI Point of Interest
WKT Well-Known Text
GIS Geographic Information System
DTNs Delay-Tolerant Networks
OppNets Opportunistic Networks
RBMM Route-Based Map Mobility Model
WSN Wireless Sensor Networks
RPGM Reference Point Group Mobility Model
RBM Random Border Mobility Model
PMM Pursue Mobility Model
228  Opportunistic Networks

12.3 STOCHASTIC MOBILITY MODEL

Mobility models are relatively easy to research under stochastic mobility [6] but show little
or no resemblance to practical scenarios. Generally, they depend on random node movement.

12.3.1 Random-Based Mobility Models

In the random-based mobility models [7, 8], mobile nodes travel independently in the network
without any constraints. The speed of nodes and destination are both randomly selected. The
random-based mobility models are very easy to implement; hence, they are used in different
types of simulation studies. These mobility models are further classified into three classes:
RWM, RWP, and RDM.

12.3.1.1 Random Walk Model

This model [9] is the oldest mobility model. It is a quite simple model based on the Brownian
Motion model. In this model, the nodes are selected randomly within a fixed range with the
direction and degree. The nodes move around the first or initial position and cannot be out
of the simulation area, and in the end the nodes return to initial position in one- and two-
dimensional space. Figure 12.2 introduces the most general case of such a network: nodes
travel around in space, and they may gain from the opportunity to forward a message when-
ever they are in random contact (given a transmission range) with another. The movement
of nodes is not dependent on others. The simulation area should be sufficiently large for the
efficient use of this mobility model.

12.3.1.2 Random Direction Model

This is a model that moves nodes to the border before changing direction and speed of the
simulation area. The main aim of this model is to remove the limitation of the RWM [10].
The probability of a node passing through the simulation is greater, as compared to the
RWM. It is simple, flexible, and easy to implement. Thus many simulations use this model.
Although, there are certain limitations of this as it considers the only direction as the main
parameter and not be a good response node in many states.

Figure 12.2  We can see the randomized opportunity ties between nodes in blue.
 Mobility Models in Opportunistic Networks  229

12.3.1.3 Random Waypoint Model

This model [11] is used in various simulations in mobile wireless ad hoc networks, and
it includes the pause time between changes in direction or location. The implementation
process is easy for this model. Several deficiencies exist in this model; for instance, as time
increases, the simulation process is attenuated, and due to spatial distribution of nodes, the
simulation process is changing from time to time. This model follows the memoryless prop-
erty; that is, it does not contain information about past behavior. These deficiencies have been
addressed by the modified version, namely the RBM and Restricted RWP. In Figure 12.3, the
pseudo code of RWP model is presented [12].

12.4 SYNTHETIC MODELS

Synthetic models [13] are the mathematical models that capture the movement of nodes by
imposing constraints like obstacles, pathways, etc. They do not rely on the random move-
ment of nodes; rather, they move in a correlated manner to capture the movement of the
nodes realistically.

12.4.1 Temporal Dependency-Based Mobility Models

The physical constraint of the mobile nodes affects the velocity of the mobile node which
gradually or abruptly changes. The temporal dependency property describes the velocity of
the nodes which are near each other at two-time instances, i.e., the current velocity of the
node is dependent upon the previous velocity. These models of mobility based on temporal

Figure 12.3  Pseudo code of RWP model.

230  Opportunistic Networks

dependence are further classified into two categories: the model of Gauss mark and the
model of smooth random mobility. Each mobile node is initially assigned a current speed and
direction in the Gauss mobility model, and the movement of each node occurs by updating
its speed and direction at a fixed speed. The time intervals are also important. Every mobile
node’s speed and direction are associated over time and are calculated upon the previous
speed and direction of the corresponding mobile node. In the smooth mobility model, the
speed of the mobile node changes gradually rather than with sudden acceleration and sharp
turns. Each node is defined by a motion vector which consists of speed and direction.

12.4.2 Spatial Dependency-Based Mobility Models

Under this group, the mobility model [14] is characterized by a spatial dependence property
in which the movement of the mobile node is associated with that of the neighboring nodes.
The Reference Point Group Mobility Model (RPGM), the Pursue Mobility Model (PMM),
and the Community Mobility Model (CMM) are known as various spatial-dependency
mobility models. In the RPGM, each mobile node in a group is randomly distributed around
the predefined reference point, which allows individual random movement of nodes along
with the group motion behavior. Each group of nodes has a logical center that determines
the entire movement of the group through the group motion vector. As the reference point
moves from the first point to the second point, the position of the actual reference point is
modified according to the logical center of the group. The motion vector can be designed on
the basis of predefined paths, or it can be selected at random. If the community motion vec-
tor has passed the checkpoint, the next checkpoint is again computed and transferred to that
checkpoint. For example, a group of soldiers march toward their enemy by forming a line.
A minor modification of the mobility model is where a set of mobile nodes are placed in a
single-file line and are allowed to follow one another about their initial position. The PMM is
another spatially dependent-based mobility model, in which mobile nodes attempt to track a
particular target. It can be used for law enforcement and signal source tracking, as the nodes
in the model try to capture the single target node [15].

12.5 GEOGRAPHICAL RESTRICTION-BASED MOBILITY MODEL

In random mobility models, mobile nodes are free to move anywhere within the simulation
area without any geographical restrictions like environmental obstacles such as pathways,
streets, buildings, etc. Geographical-restriction-based mobility models are models that are
bounded by environmental restrictions [16].

12.5.1 Pathway Mobility Model

The Pathway mobility model [17] is a geographical-restriction-based mobility model where
the model is bounded by pathways. In this model, the map is predefined for the simulation
area, and it models the map as a random graph where vertices in the graph represent the
buildings, and edges represent pathways.

12.5.2 Obstacle Mobility Model

The Obstacle mobility model [18] is another geographical-restriction-based mobility
model that tries to model the movement of the node considering the environmental obsta-
cles. Environmental obstacles affect the movement pattern of the nodes, as when the node
 Mobility Models in Opportunistic Networks  231

encounters the obstacle, it has to change its path. Therefore, obstacles play a significant part
in modeling the mobility of the node [19].

12.5.3 Manhattan Mobility Model

The Manhattan mobility model [20] is also a geographical-restriction-based mobility model
which uses predefined grid topology for the movement of the nodes. It models the movement
pattern of the nodes in well-organized streets of urban areas. It moves in a horizontal and
vertical direction by using a probabilistic approach to move straight, left, and right on the
street.

12.5.4 City Segment Model

The mobility model of the City Segment mimics the realistic movement of the node in a town
segment. In the mobility model, the mobile nodes follow the predefined route and are limited,
as in real scenarios, by obstacles and traffic regulation. Each mobile node begins its movement
from a defined point on the street toward the randomly chosen destination through the
shortest path. After reaching the destination, it stays at that location for a specific pause time,
and again it chooses a random destination for the movement and repeats the process until
the simulation ends. The City Section mobility model presents the realistic movement of the
node by restricting the movement pattern, that is, following the predefined path like in real
scenarios, as people move on a predefined path following traffic regulations.

12.6 MAP-BASED MOBILITY MODELS

In these models, using a subset of the Well-Known Text (WKT) format, map data is used. In
reality, WKT is based on the ASCII format commonly used in the programs of the Geographic
Information System (GIS) [4]. Map-based mobility models [21, 22] are based on the traffic
and user mobility patterns to evaluate and deal with networking issues in the OppNets.
Nodes in map-based mobility models move randomly on a predetermined path defined by
the map data in a well-known text format file [23–25]. In map-based mobility models, a
group of nodes can select a certain destination or point on the map, and after traveling to a
specified distance, they pause for a certain pause time and then resume their journey.

12.6.1 Route-Based Map Mobility Model

The Route-Based Map Movement (RBMM) mobility model [26] mimics the movement of
nodes based on predefined routes on the simulation maps. Nodes in this model have prede-
termined routes that they follow in the simulation map after getting the map data, but during
the movement of nodes, the model does not select the destination randomly; instead, they
select the next destination to which they are currently moving. The routes in the model have
many stop points which the nodes follow while traveling; they stop for a specified pause time
on the stop points to reach the destination with the shortest path for modeling the movement
pattern.

12.6.2 Rush Hour (Human) Traffic Model

The Rush Hour Traffic model [27] deals with the traffic condition of rush hour. The Rush
Hour Traffic model focuses on high traffic area and choice of destination, as people tend to
232  Opportunistic Networks

travel to nearby places rather than going to further destinations. The algorithm is based on
the exponential distribution.

12.6.3 Working Day Movement Model

The WDM [28] was proposed to capture movement patterns of real-life scenarios. To deal
with inter-contact time and contact time distributions that are contained in the traces of
real-world measurement experiments, this model was proposed for DTNs and OppNets.
With four sub-models, the WDM is developed, representing the work routine that most indi-
viduals follow every day and repeatedly. Home activity, workplace activity, evening activity,
and transport activity are these sub-models. The nodes move on a map containing all the
positions of activities at home, office, and evening, and identifying spaces and routes for the
nodes to move. The WDM tries to model each scenario that people follow in everyday life.
In Figure 12.4, the pseudo code of WDM model is presented [12].

Figure 12.4  Pseudo code of WDM model.

 Mobility Models in Opportunistic Networks  233

12.6.4 Shortest Path Map-Based Movement Model

The Shortest Path Map-Based Movement (SPMBM) mobility model [29] proposed for DTNs
is based on the shortest path available within the simulation map scenario. In this model,
rather than moving randomly around the simulation map, the shortest path is chosen over
the available paths between two random nodes and points of interest (POIs) from the simula-
tion map, as people mostly use the shortest path to travel. POIs are the places on the simula-
tion map to which nodes within the group travel.
In the SPMBM, all the nodes are placed randomly on the simulation map, and then nodes
travel to a certain destination in the simulation map following the Dijkstra algorithm to
discover the shortest path from the available paths on the simulation map. After reaching
a specified destination, the nodes pause for a certain period of time and again move to a
newly selected destination through the shortest path. Map-based mobility models generally
model movement patterns based on the map, the routes people follow, the traffic pattern, the
working activities pattern, etc. These mobility models are useful in modeling various real-life
scenarios, but the map-based mobility models have certain limitations in that the paths in
the model are predefined, which, in some cases, is not possible in real life. In Figure 12.5, the
pseudo-code of the SPMBM model is presented [12].

12.6.5 Social Network-Based Mobility Models

The social network-based mobility models [30–32] capture the behavior of movement based
on human decisions and the social nature of humans, or the entities just like the social

Figure 12.5  Pseudo-code of the SPMBM model.

234  Opportunistic Networks

behavior of humans in a battlefield, during disaster relief, etc. They rely upon the structure of
the relationships among the individuals.

12.6.6 Community-Based Mobility Models

The CMM is a social network-based mobility model which mimics the movement pattern
based on social network [33]. The CMM mobility model groups the node according to the
community to which they belong. Nodes with the same community are grouped as friends,
and nodes that belong to different communities are grouped as non-friends. Initially, a cell
is allotted to each community where they share a social link between the friends’ and non-
friends’ communities for the movement of the nodes in the network.

12.6.7 Social Network Models

In this model, artificial users periodically move between abstract locations. A set of abstract
locations, which they have to visit at a fixed time, is given to the users. In [22], the author
proposed a social network model is based on social network theory as it depends upon
human relationships. The model allows several hosts to be grouped based on the social
relationships among them and mapped to a topographical space. It models the social net-
work through a weighted graph of the relationships. Social network models are based on
the social relationship between the nodes, which affects the movement pattern of the node
in the network; hence, it is necessary to model social behavior based on social network
theory [34].

12.7 SIMULATION ANALYSIS

The Simulator for the Opportunistic Network Environment (ONE) [12] is a simulation tool
commonly used by researchers studying DTNs and OppNets. Its salient features are listed as
follows:

• The ONE simulator is a tool for researchers to carry out research-based simulations
related to DTNs, WSN, and OppNets
• It develops and tests different movement models for nodes
• It adds message routing capabilities between nodes
• It uses JUnit and the GUI to debug the protocols being developed as well as visualize
them
• It generates results and statistics for tests effortlessly.

The ONE simulator’s default movement mobility model settings are shown in Figure 12.6.
The map-based movement model requires that settings define the files being used to construct
the movement models. It can be seen that the four map files used are essentially defining the
paths and places of the real world that are being simulated. The reader can define their own
map files in the WKT format. This will be covered later in more detail. One can also define
the area of the world that they want for their simulations as well as the warmup time, which
is the time duration for which hosts have moved around in the defined area until the onset
of the real simulation. The router classes are connected. This is clearly shown in Figure 12.7.
One class inherits another class, and subsequent inheritance takes place.
In Figure 12.8, the interaction of simulator files is shown. The simulation parameters are
listed in Table 12.2.
 Mobility Models in Opportunistic Networks  235

Figure 12.6  Default movement mobility model settings.

Figure 12.7  Inheritance between routers.

Text UI Movement Model

Simulation Message Message

Start Generator

GUI

Interface
Host
Connection
Maps

SimScenario Router
Routes
From File

Reports Output
Listener
Settings

World

Figure 12.8  Interaction flow of ONE Simulator files.

236  Opportunistic Networks

Table 12.2  Simulation parameters.

Parameter Value

Duration of simulation 43200s

Dimension 4500 × 3200 m
Nodes count in simulation 130–455
Velocity of nodes 0.5–1.5 m/s
Range 10 m
Speed 25 Kbytes/s
Size of message 1 Kbytes
Number of host groups 4
Group waiting time 0–120s
Group speed 0.5–1.5 m/s
Number of events 1
Events interval 25–35s
Map Helsinki Map
GUI Under lay image offset (64, 20)
GUI Under lay image scale 4.75
GUI Under lay image rotate –0.015
GUI Event log panel 30
GUI Under lay Image helsinki underlay
Report warmup time 0s
Movement model warmup 1000s
Movement model random number generator seed 1

12.7.1 Testing Tool
In this subsection, we show the flow of simulation files that are required for each simulation.
Figure 12.9 represents init model flow. The DTNSim file is the main file of ONE simulation.
This initModel() is directly linked to the main file and respective setting files. EventQueue()
and createHosts() are called by initModel() for simulation. Figures 12.10 and 12.11 repre-
sent runtime flow and update flow, respectively. The world() is called when the simulation is
running. The router update() is one of the main update functions for routing. Figure 12.12
represents the execution window of the ONE simulator at an instance. The Helsinki map
model is used in the simulation.

12.7.2 Simulation Results
In this subsection, the PRoPHET protocol [35] is selected for simulation. The reason behind
the PRoPHET routing scheme selection is better delivery probability.
PRoPHET’s performance is simulated and compared against that of benchmark mobility
models SPMBM, RBMM, and RWP mobility models using the ONE simulator. The per-
formance metrics for simulating the mobility models for the PRoPHET routing scheme are
delivery probability and average latency. Figure 12.13 shows that RBMM offers better distri-
bution than the RWP and SPMBM with the PRoPHET routing protocol. This means that the
delivered packets are high in the RBMM mobility model, while the number of nodes in the
network is smaller. This is good for OppNets since, with a limited number of nodes, the net-
work is first generated and then extended. If the network is sparse, better delivery probability
 Mobility Models in Opportunistic Networks  237

Figure 12.9  initModel() flow chart.

Figure 12.10  runSim flow() chart.

can also be done. Figure 12.14 shows PRoPHET’s performance when the delay is considered
as the performance metric.
The buffer capacity is varied and the effect of this adjustment is investigated on the likeli-
hood of delivery. In Figure 12.15, the output of buffer policies is captured. It is found that
with increasing buffer power, the likelihood of delivery is also improved. Because of a node’s
238  Opportunistic Networks

Figure 12.11  Delivery Probability vs TTL.

Figure 12.12  update() flow chart.

buffer capacity, the more messages stored in that buffer are increased, the more messages that
are sent to the receiver node. The buffer capacity is varied, and the impact of this adjustment
is examined on the average latency. In Figure 12.16, the effects of the buffer policies are cap-
tured. The TTL is diverse and the effect on the latency of this transition is observed. Figure
12.17 shows the results of delivery probability versus time to live. Figure 12.18 displays
the results of average latency versus time to live. It is found that the average latency often
increases when the TTL is boosted. This is because a substantial TTL value often increases
the message’s stay in the buffer of the node.
 Mobility Models in Opportunistic Networks  239

Figure 12.13  ONE simulator: execution screenshot at an instance.

Figure 12.14  Delivery probability vs. number of nodes.

Figure 12.15  Average latency vs. number of nodes.

240  Opportunistic Networks

Figure 12.16  Delivery probability vs. buffer size.

Figure 12.17  Average latency vs. buffer size.

Figure 12.18  Average latency vs. TTL.

12.8 CONCLUSION

In this chapter, the mobility models were discussed for opportunistic networks in detail and
summarized according to simulation outcomes. The simulation study showed that the best
mobility model is dependent on regular routes. We conducted a performance simulation of
PRoPHET using the RWP, SPMBM, and RBMM models. The results of the simulation indi-
cate that PRoPHET produces better performance in the case of the RBMM model.
Humans are found in our research as fundamental components of many opportunistic
applications. We should therefore bring greater emphasis on balancing distinct human behav-
ior with other problems of modeling. Since contact times are a microscopic human mobility
property, the research built around them has tremendous significance in realistic modeling
of mobility, especially when one departs from the exponential assumption or introduces a
small amount of heterogeneity. In mobility modeling, it is believed that human mobility and
subsequent contacts occur by purpose and position. Secondly, social links between nodes
 Mobility Models in Opportunistic Networks  241

direct a node to choose the destination and the timing of a mobility journey, while the loca-
tion provides guidance on the direction to be followed. This produces a very complex system
of touch that is not easily visible at the inter-contact level.
Therefore, to capture the array of node interrelations in a more tractable way, a more
abstract and macroscopic view of mobility is needed. Greater emphasis should be placed on
integrating various human behaviors together with modeling problems in future researchers.

REFERENCES

[1] M. Conti et al., “From Opportunistic Networks to Opportunistic Computing,” IEEE Commun.
Mag., vol. 48, 2010, pp. 126–139.
[2] C. Boldrini, M. Conti, and A. Passarella, “Exploiting User’s Social Relations to Forward Data
in Opportunistic Networks: The HiBOp Solution,” Pervasive Mob. Comput., vol. 4, 2008, pp.
633–657.
[3] M. C. Gonzalez, C. A. Hidalgo, and A.-L. Barabasi, “Understanding Individual Human Mobility
Patterns,” Nature, vol. 453, 2008, pp. 779–782.
[4] C. Song et al., “Modeling the Scaling Properties of Human Mobility,” Nature Phys., vol. 6, 2010,
pp. 818–823.
[5] P. Hui et al., “Pocket Switched Networks and Human Mobility in Conference Environments,”
Proceedings of the 2005 ACM SIGCOMM Workshop on Delay-Tolerant Networking, 2005, pp.
244–251.
[6] D. Brockmann, L. Hufnagel, and T. Geisel, “The Scaling Laws of Human Travel,” Nature, vol.
439, 2006, pp. 462–465.
[7] A. Chaintreau et al., “Impact of Human Mobility on Opportunistic Forwarding Algorithms,”
IEEE Trans. Mobile Comp., vol. 6, 2007, pp. 606–620.
[8] B. D. Walker, T. C. Clancy, and J. K. Glenn, “Using Localized Random Walks to Model Delay-
Tolerant Networks,” in Military Communications Conference, 2008. MILCOM 2008, 2008, pp.
1–7. IEEE.
[9] T. Karagiannis et al., “Power Law and Exponential Decay of Intercontact Times between Mobile
Devices,”IEEE Trans. Mobile Comp., vol. 9, 2010, pp. 1377–1390.
[10] C. Song et al., “Limits of Predictability in Human Mobility,” Science, vol. 327, 2010, pp.
1018–1021.
[11] J. Ghosh, S. J. Philip, and C. Qiao, “Sociological Orbit Aware Location Approximation and
Routing (SOLAR) in MANET,” Ad-Hoc Net., vol. 5, 2007, pp. 189–209.
[12] A. Kernen, J. Ott, and T. Krkkinen, “The ONE Simulator for DTN Protocol Evaluation,” in
Proceedings of the 2nd International Conference on Simulation Tools and Techniques, 2009
(March), pp. 1–10.
[13] W. J. Hsu et al., “Modeling Time-Variant User Mobility in Wireless Mobile Networks,” IEEE
INFOCOM 2007, 2007, pp. 758–766.
[14] K. Lee et al., “SLAW: A New Mobility Model for Human Walks,” IEEE INFOCOM 2009, 2009,
pp. 855–863.
[15] A. Mei and J. Stefa, “SWIM: A Simple Model to Generate Small Mobile Worlds,” IEEE
INFOCOM 2009, 2009, pp. 2106–2113.
[16] I. Rhee, K. Lee, S. Hong, S. J. Kim, and S. Chong, “Demystifying the Levy-Walk
Nature of Human Walks,” Technical Report, NCSU, http://netsrv.csc.ncsu.edu/export/
DemystifyingLevyWalkPatterns.pdf, 2008.
[17] Q. Zheng et al., “Agenda Driven Mobility Modeling,” Int’l. J. Ad Hoc Ubiquitous Comput., 5,
2010, pp. 22–36.
[18] F. Ekman et al., “Working Day Movement Model,” Proceedings of the 1st ACM SIGMOBILE
Workshop Mobility Models, 2008, pp. 33–40.
[19] M. Musolesi and C. Mascolo, “Designing Mobility Models Based on Social Network Theory,”
SIGMOBILE Mob. Comp. Commun. Rev., 11, 2007, pp. 59–70.
242  Opportunistic Networks

[20] C. Boldrini and A. Passarella, “HCMM: Modeling Spatial and Temporal Properties of Human
Mobility Driven by User’s Social Relationships,” Computer Commun., 33, 2010, pp. 1056–1074.
[21] V. Borrel et al., “SIMPS: Using Sociology for Personal Mobility,” IEEE/ACM Trans. Net., 17,
2009, pp. 831–842.
[22] S. Yang et al., “Using Social Network Theory for Modeling Human Mobility,” IEEE Netw., 24,
2010, pp. 6–13.
[23] D. Thakore and S. Biswas, “Routing with Persistent Link Modeling in Intermittently Connected
Wireless Networks,” in Military Communications Conference, 2005. MILCOM 2005, 2005
(October), pp. 461–467. IEEE.
[24] F. Warthman, “Delay Tolerant Network – A Tutorial,” version 1.1, May 2003.
[25] R. S. Mangrulkar and M. Atique, “Routing Protocol for Delay Tolerant Network: A Survey
and Comparison,” in 2010 IEEE International Conference on Communication Control and
Computing Technologies (ICCCCT), 2010 (October), pp. 210–215. IEEE.
[26] D. Fischer, K. Herrmann, and K. Rothermel, “GeSoMo: A General Social Mobility Model for
Delay Tolerant Networks,” 2010 IEEE 7th International Conference on Mobile Ad Hoc Sensor
System, 2010, pp. 99–108.
[27] A. Bar-Noy, I. Kessler, and M. Sidi, “Mobile Users: To Update or Not to Update?” Wirel. Netw.,
1(2), 1995, pp. 175–185.
[28] M. Grossglauser and D. Tse, “Mobility Increases the Capacity of Ad-hoc Wireless Networks,”
in INFOCOM 2001, Twentieth Annual Joint Conference of the IEEE Computer and
Communications Societies. Proceedings. IEEE, Anchorage, AK, vol. 3, 2001, pp. 1360–1369.
IEEE.
[29] W. J. Hsu, T. Spyropoulos, K. Psounis, and A. Helmy, “Modeling Time-Variant User Mobility
in Wireless Mobile Networks,” in INFOCOM 2007, 26th IEEE International Conference on
Computer Communications. IEEE, Barcelona, Spain, 2007 (May), pp. 758–766.
[30] W. J. Hsu, T. Spyropoulos, K. Psounis, and A. Helmy, “Modeling Spatial and Temporal
Dependencies of User Mobility in Wireless Mobile Networks,” IEEE/ACM Trans.Network
(ToN), 17(5), 2009, 1564–1577.
[31] W. Gao, Q. Li, B. Zhao, and G. Cao, “Multicasting in Delay Tolerant Networks: A Social
Network Perspective,” in Proceedings of the Tenth ACM International Symposium on Mobile
Ad Hoc Networking and Computing, 2009, pp. 299–308. ACM.
[32] T. Hossmann, T. Spyropoulos, and F. Legendre, “Know Thy Neighbor: Towards Optimal
Mapping of Contacts to Social Graphs for DTN Routing,” in INFOCOM, 2010 Proceedings
IEEE, 2010, pp. 1–9. IEEE.
[33] J. Lakkakorpi, M. Pitknen, and J. Ott, “Adaptive Routing in Mobile Opportunistic Networks,” in
Proceedings of the 13th ACM International Conference on Modeling, Analysis, and Simulation
of Wireless and Mobile Systems, 2010 (October, pp. 101–109. ACM.
[34] B. D. Walker, J. K. Glenn, and T. C. Clancy, “Analysis of Simple Counting Protocols for Delay-
Tolerant Networks,” in Proceedings of the Second ACM Workshop on Challenged Networks,
2007, pp. 19–26. ACM.
[35] A. Lindgren, A. Doria, E. Davies, and S. Grasic, (2012, August), “Probabilistic Routing Protocol
for Intermittently Connected Networks,” IETF 6693.
Chapter 13

Opportunistic Routing in Mobile Networks

Zehua Wang
University of British Columbia, Vancouver, Canada

Yuanzhu Chen, and Cheng Li

Memorial University of Newfoundland, St. John’s, Canada

CONTENTS

13.1 Introduction............................................................................................................. 244

13.1.1 Mobile Ad Hoc Networks (MANETs) .......................................................245
13.1.2 Extremely Opportunistic Routing............................................................... 245
13.2 Related Work and Motivation................................................................................. 246
13.2.1 Recent Work Related to Opportunistic Data Forwarding........................... 246
13.2.1.1  Protocols Based on Opportunistic Data Forwarding.................. 246
13.2.1.2  Other Variants of Opportunistic Forwarding............................. 250
13.2.1.3  Scenarios Suitable for Opportunistic Forwarding....................... 251
13.2.1.4  Performance Modeling for Opportunistic Forwarding............... 251
13.2.2 Routing Algorithms Review for Opportunistic Data Forwarding in
MANETs ...................................................................................................252
13.2.2.1  Timing Strategy in Routing Protocols......................................... 252
13.2.2.2  Fundamental Algorithms – Link State and Distance Vector........ 253
13.2.2.3  Tree-Based Routing Protocols Derived from the Internet............ 253
13.2.2.4 Suitability of Existing Routing Protocols for Opportunistic
Data Forwarding in MANETs �������������������������������������������������� 254
13.2.3 Motivation and Framework........................................................................ 254
13.2.3.1  Objectives and Challenges .........................................................254
13.2.3.2  CORMAN Fundamentals .........................................................255
13.3 Proactive Source Routing – PSR............................................................................... 256
13.3.1 Design of PSR............................................................................................. 256
13.3.1.1  Route Update............................................................................. 256
13.3.1.2  Neighborhood Trimming........................................................... 257
13.3.1.3  Streamlined Differential Update................................................. 258
13.3.2 Implementation.......................................................................................... 259
13.3.2.1  Routing and Neighborhood Update Algorithm.......................... 259
13.3.2.2  Algorithms for Transformation of Tree Structures...................... 260
13.3.2.3  Implementation of Reconstruction in Differential Update.......... 261
13.4 Large-Scale Routing Update and Small-Scale Retransmission ..................................265
13.4.1 Large-Scale Live Update............................................................................. 265
13.4.2 Small-Scale Retransmission......................................................................... 267
13.4.2.1  Considerations of Small-Scale Retransmission........................... 267

243
244  Opportunistic Networks

13.4.2.2  Design of Small-Scale Retransmission........................................ 268

13.4.2.3  Algorithm and Scoring Function................................................ 269
13.5 Performance Evaluation........................................................................................... 273
13.5.1 Performance Study of PSR.......................................................................... 274
13.5.1.1  Experiment Settings................................................................... 274
13.5.1.2  TCP with Node Density............................................................. 275
13.5.1.3  TCP with Velocity...................................................................... 276
13.5.1.4  UDP with Density...................................................................... 279
13.5.1.5  UDP with Velocity...................................................................... 281
13.5.2 Effectiveness Study of Small-Scale Retransmission...................................... 281
13.5.2.1  Experiment Settings................................................................... 282
13.5.2.2  Performance versus Network Dimension................................... 283
13.5.2.3  Performance versus Velocity....................................................... 286
13.5.3 Overall Performance of CORMAN............................................................ 286
13.5.3.1  Experiment Settings................................................................... 287
13.5.3.2  Performance versus Network Dimension................................... 288
13.5.3.3  Performance versus Velocity....................................................... 290
13.6 Concluding Remarks............................................................................................... 292
13.6.1 Conclusions................................................................................................ 292
13.6.2 Discussion.................................................................................................. 293
13.6.2.1  Proactive Source Routing Related.............................................. 293
13.6.2.2  About Large-Scale Live Update and Small-Scale
Retransmission........................................................................... 294
13.6.3 Future Work............................................................................................... 295
References........................................................................................................................ 295

13.1 INTRODUCTION

The multi-hop wireless network has drawn a great deal of attention in the research com-
munity. Within the long period after the network was proposed, the routing and forward-
ing operations in the network have remained quite similar to those in the multi-hop wired
network or the Internet. However, all the data transmission over the wireless medium in the
wireless networks is by broadcasting in nature, which is different from the Internet. Because
of the broadcast nature, many opportunities based on overhearing can be used to enhance the
data transmission ability in the wireless network. ExOR is the first practical data forwarding
scheme which tries to promote data transmission ability by utilizing the broadcast nature
in wireless mesh networks, and opportunistic data forwarding becomes a well-known term
given by ExOR to name this kind of new data forwarding scheme. The basic idea in ExOR
has triggered a great deal of derivations. However, almost all these derivations are proposed
for wireless mesh networks or require the positioning service to support opportunistic data
forwarding in mobile ad hoc networks (MANETs). In this chapter, we propose a series of solu-
tions to implement opportunistic data forwarding in more general MANETs, which is called
Cooperative Opportunistic Routing in Mobile Ad-hoc Networks (CORMAN). CORMAN
includes three following important components. First, a new lightweight proactive source
routing (PSR) scheme is proposed to provide source routing information in MANETs for
both opportunistic data forwarding and traditional IP forwarding. Second, we analyze
and evaluate the topology change with mathematical model and propose large-scale live
update to update routing information more quickly with no extra communication overhead.
 Opportunistic Routing in Mobile Networks  245

Third, we propose that small-scale retransmission utilizes the broadcast nature one step fur-
ther than ExOR, and furthermore, it helps us to enhance the efficiency and robustness of the
opportunistic data forwarding in MANETs.

13.1.1 Mobile Ad Hoc Networks (MANETs)

A mobile ad hoc network (MANET) is a wireless communication network, where nodes that
are not within direct transmission range of each other will require other nodes to forward
data. It can operate without existing infrastructure, supports mobile users, and falls under the
general scope of multi-hop wireless networking. Such a networking paradigm was originated
from the needs in battlefield communications, emergency operations, search and rescue, and
disaster relief operations. Later, it found civilian applications such as community networks.
A great deal of research results have been published since its early days in the 1980s [1]. The
most salient research challenges in this area include end-to-end data transfer, link access con-
trol, security, and providing support for real-time multimedia streaming.
The network layer has received the most attention when working on MANETs. As a result,
abundant routing protocols in such a network with differing objectives and for various spe-
cific needs have been proposed [2]. In fact, the two most important operations at the network
layer, i.e., data forwarding and routing, are distinct concepts. Data forwarding regulates how
packets are taken from one link and put on another. Routing determines what path a data
packet should follow from the source node to the destination. The latter essentially provides
the former with control input.
Routing protocols for MANETs can be classified based on a variety of factors. The most
important distinction is the way in which routing information is maintained or updated. On
the one hand, a protocol can be proactive, requiring network nodes to always maintain valid
routes to all destinations. In this case, the protocol is also known as table driven. Destination-
Sequenced Distance Distance-Vector (DSDV) [3] and Optimized Link State Routing (OLSR)
[4] are two proactive routing protocols. A reactive protocol, on the other hand, is used when
a node receives data from the upper layer for a given destination and must first figure out
how to get there if it does not always possess the up-to-date routing information. Also known
as on-demand, the reactive protocol is used in Dynamic Source Routing (DSR) [5] and Ad
Hoc On-Demand Distance Vector (AODV) [6].

13.1.2 Extremely Opportunistic Routing

Despite the efforts in routing in ad hoc networks, data forwarding, in contrast, follows pretty
much the same paradigm as in IP forwarding in the Internet. IP forwarding was originally
designed for multi-hop wired networks, where one packet transmission can be received only
by nodes attached to the same cable. For the case of modern Ethernet, an IP packet is trans-
mitted at one end of the Ethernet cable and received at the other. However, in wireless net-
works, when a packet is transmitted over a physical channel, it can be detected by all other
nodes within the transmission range on that channel. For the most part of the research his-
tory, overhearing a packet not intended for the receiving node had been considered as com-
pletely negative, i.e., interference. Thus, the goal of research in wireless networking was to
make wireless links as good as wired ones. Unfortunately, this ignores the inherent nature of
broadcasting of wireless communication links. For mobile ad hoc networks to truly succeed
beyond labs and testbeds, we must tame and utilize its broadcasting nature rather than fight
it. Cooperative communication is an effective approach to achieve such a goal.
246  Opportunistic Networks

The concept of cooperative communication was initially put forward by Cover and El
Gamal [7], who studied the information theoretic properties of relay channels. More recent
progress on this subject started to proliferate in the early 2000s [8]. In cooperative commu-
nication at the physical layer, multiple nodes overhearing the same packet may transmit it
together as a virtual multiple-antenna transmitter. With enhanced digital signal-processing
capabilities on the receiver side, the packet is more likely to be decoded. Yet, little research
had been done on cooperative commutation at the link layer and above until ExOR (Extreme
Opportunistic Routing) [9]. ExOR is a milestone piece of work in this area, and it is an
elegant way to utilize the broadcasting nature of wireless links to achieve cooperative com-
munication at the link and network layers of static multi-hop wireless networks. Therefore,
in our research, we further extend the scenarios that the idea behind ExOR can be used,
dubbed as Cooperative Opportunistic Routing in Mobile Ad-hoc Networks (CORMAN).

13.2 RELATED WORK AND MOTIVATION

In this part of the chapter, we review recent work in two related fields. We first review oppor-
tunistic data forwarding, including its ancestors and derivations, the related math models
built for it, and its effectiveness studies. We then review the importance of routing protocols,
and hence we will review some routing protocols in the second part of this chapter. After
the review of related work, we will highlight the motivation and introduce the framework
in our research.

13.2.1 Recent Work Related to Opportunistic Data Forwarding

This part provides an overview of opportunistic data forwarding in multi-hop wireless net-
works by explaining and comparing existed ones. They may include many trails for different
coordinate protocols, metrics, network types and so on. The discussion in this part is organized
as follows. Section 13.2.1.1 will present and compare the mechanisms used to realize oppor-
tunistic data forwarding. Section 13.2.1.2 will list other derivations in opportunistic data for-
warding. Section 13.2.1.3 will review the question of when the opportunistic data forwarding
can achieve better performance than traditional IP forwarding in a multi-hop network. Section
13.2.1.4 contains the mathematical models that can help us to analyze the performance of
opportunistic data forwarding and understand how to maximize such performances.

13.2.1.1 Protocols Based on Opportunistic Data Forwarding

In recent years, dozens of opportunistic data forwarding protocols have been proposed. Many
of them are significant in this area. In this section, we present the challenges in opportunistic
data forwarding and how to address them. First, we need to know where the advantages of
opportunistic data forwarding comes from. The two most common scenarios that exist in
wireless data transmission are shown in Figures 13.1 and 13.2 [9].
In Figure 13.1, the source node src transmits a packet to the destination node dst. By the
topology and link delivery probabilities shown in the figure, we know that the probability of
at least one intermediate node receiving the data from src is 1 − (1 − 0.1)100, which is greater
than 10%, the probability of a particular node receiving it. Furthermore, any intermediate
node can forward the packet. It is one reason why a better performance may be achieved in
opportunistic data forwarding.
Figure 13.2 presents another possibility that one transmission may reach a node which is
closer to the destination than the particular next hop in traditional IP forwarding. Assume in
 Opportunistic Routing in Mobile Networks  247

2
src dst

100

Figure 13.1  Topology and deliver probabilities as an example network.

A B C D E

Figure 13.2  Best node can be used in opportunistic data forwarding.

traditional IP forwarding, the next hop for a packet sent from A is B, and node C overhears
the packet by opportunity. In the traditional approach, this is taken as the interference on
node C, and the overheard packet will be discarded. Meanwhile, node B will transmit it
to C again. What makes it worse is that if node B cannot decode the packet successfully,
even though node C could properly decode the packet, the benefit cannot be utilized in
the IP forwarding scheme, and node A should transmit the packet again until B receives
it. Opportunistic data forwarding can postpone the decision to choose the best forwarder
which may be far away from the transmitter but receives the packet by a certain opportu-
nity. Note that some situations may diminish the benefits of opportunistic data forwarding
because the next hop is not determined and any node receiving the data could be an actual
relay. Hence the duplicated transmission from many intermediate nodes should be avoid. If
the cooperation protocol cannot guarantee that only one forwarder candidate will forward
the data, duplicated transmission would decrease the performance. Moreover, if the coor-
dination mechanism is quite resource consuming, this protocol will be unusable. Therefore,
we believe that the coordination protocol, which serves opportunistic data forwarding, is
the key point. How to design a coordination protocol for forward candidates will be a
dominant issue to make opportunistic data forwarding more efficient and more practical. As
far as we know, two protocols for coordination, introduced next, are significant for oppor-
tunistic data forwarding.

13.2.1.1.1 Multiple Handshake
Larsson [10] proposes a four-way handshake approach as the coordination protocol in his
Selection Diversity Forwarding (SDF). In SDF, if a node has a packet to transmit, it just broad-
casts the packet to every neighbor. Then, every neighbor that received the packet successfully
will send back an acknowledgement (ACK) with their local information to the transmitter.
The transmitter makes a decision based on the ACKs and sends a forwarding order (FO) to
the best forwarder candidate. Once the selected relay node receives the FO, it will send the
forwarding order ACK (FOA) back to the transmitter and then proceed to data forwarding.
This process continues until the final destination is reached. As a piece of pioneering work on
opportunistic data forwarding, Larsson has made a significant contribution. However, people
realize that two problems exist in Larsson’s work. One is that the ACKs and FOA may be
248  Opportunistic Networks

lost in the wireless environment, and the loss of either one will lead to unnecessary retrans-
missions. The other is that such a gossiping mechanism wastes a great deal of resources and
introduces further delay. By this consideration, Rozner et al. [11] explored the approaches to
make multiple- handshake coordination tolerate ACK loss and reduce the delay. Rozner et al.
used selective ACKs to address ACK losses and minimize useless retransmissions, and used
piggybacked ACKs and ACK compression to reduce ACK overhead. With selective ACKs, a
single ACK loss will not trigger retransmissions because subsequent ACKs for later packets
will guarantee that all packets have been received recently. In particular here, piggybacked
ACKs are constructed by including acknowledgment information to a data frame, and when
the packet is transmitted. The downstream nodes and upstream transmitter should decode
the data and ACK information from such a packet separately. Another solution based on
multiple-handshake is presented in Kurth et al. [12]. It suggests that a lower data rate can
be used to transmit ACKs to enhance their reliability, and it also explores the transmitter
diversity, which follows the same idea as receiver diversity. Naghshvar and Javidi [13] also
use multiple-handshake to coordinate all forwarder candidates. They propose a new metric,
which is the backlogs from all downstream links, to achieve better performance. However,
Naghshvar and Javidi assume that the ACKs are loss free.

13.2.1.1.2 Route-Prioritized Contention
Route-prioritized contention is the other type of coordination protocols. In route-prioritized
contention, all forwarding candidates will follow a given order to content the wireless
medium, and the best node that receives the packet should grab the medium first. Therefore,
the broadcast nature can be explored.
This approach is proposed by Biswas and Morris [9], called ExOR. ExOR is a milestone
for opportunistic data forwarding for three reasons: (1) ExOR uses piggybacked ACK to tell
the lower-priority nodes that the packet has been forwarded by higher-priority nodes. (2) The
overhead in ExOR is small, containing only the ExOR header and the metric updating infor-
mation. (3) Although it cannot completely avoid duplications, it can do so to a high degree.
Now, we try to explain this coordinate protocol in detail. The header of ExOR is shown
in Figure 13.3. In the figure, Ver, HdrLen, PatloadLen, Checksum, and Payload are self-
explanatory. In particular, to enhance the efficiency, ExOR delivers packets by batch, and
each batch contains a number of packets. The quantity of batches is recorded by the BatchSz

Figure 13.3  ExOR header.

 Opportunistic Routing in Mobile Networks  249

in the header, so the BatchSz of every packet in the same batch should be identical. The data
transmission is based on batch, which means the source will flush all packets in the same
batch into the network. Once all packets in the same batch are received by the destination
(in ExOR, when the destination received over 90% of the total packets in the same batch),
the next batch should be transmitted. The Batch ID in the header identifies which batch the
current packet belongs to, and the PktNum is the index of the packet in the current batch.
Forwarder List contains the forwarder candidates which are selected and sorted by source
node when a batch is constructed. As a result, the Forwarder List should remain the same
for all packets in one batch. If a node contained in the Forwarder List overhears a packet
and the node’s priority is higher than the transmitter’s, the packet should be buffered in the
node. In the forwarding step, to make a higher-priority node send buffered packets first,
waiting time is calculated by every node based on its position in the Forwarder List. The time
may be updated by the Transmission Tracker, which records the transmit rate and the quan-
tity of packets that need transmit from current transmitter. To support the calculation, the
FragNum and FragSz are added to the header. Here, a Fragment is a series of packets heard
by the node with ID ForwarderNum. FragSz is the quantity of packets that the current node
should transmit when its turn comes, and FragNum is the index of this packet in the current
Fragment. When a node receives a packet for the first time, the Batch Map maintained by the
node should be updated. For every packet, the highest priority forwarder that has received
it should be recorded in the Batch Map. A copy of this Batch Map, which is maintained on
the node, will be added to the header of the packet to be transmitted. Hence, when lower-
priority nodes overhear a packet from a higher-priority node, they will continue to merge the
Batch Map maintained by themselves. At the same time, they will look up the Batch ID and
PktNum in their buffers, and once a same packet is found, the transmission of this packet
would be canceled. By looking up the Batch Map, many duplications can be avoided. All nec-
essary information for coordinating forwarder candidates is integrated in the ExOR header,
thus no time is wasted for gossiping between nodes.
Many papers were published after ExOR, such as Zeng et al. [14], Yang et al. [15], and
Zhong et al. [16], and they use nearly the same coordinate protocol to make trails on other
directions. However, ExOR has its disadvantage. Because the forwarding timer is always ini-
tialized according to the node’s priority in the Forwarder List, the nodes far from destination
will always wait for a long time. It quite constrains us from exploring the spatial reuse in the
multi-hop network. Furthermore, ExOR is quite suitable for unicast, but in multicast, it may
not perform well, Chachulski et al. [17]. At last, piggybacked ACKs may lose, so duplicated
transmission may happen.
It is worth mentioning that Chachulski et al. [17] proposed MORE (Mac-independent
opportunistic routing) which is an opportunistic forwarding approach with spatial reuse,
and it also belongs to the Route-Prioritized Contention scope. As far as we know, it is the
first paper that uses network coding to realize opportunistic data forwarding. Its motiva-
tion for using network coding is that it wants to explore the spatial reuse in ExOR, without
duplicated transmission. Recall that in ExOR, the Forwarder List is generated by the source
node, and the nodes which are far from the destination will wait for a long time, even though
some of these nodes can transmit (or receive) some packets to (or from) some other nodes
without much interference on the forwarding taking place far away. The reason why ExOR
cannot operate like a pipeline is that for each packet we cannot tell how far it has already
been transmitted in the last hop. To enable pipeline opportunistic data forwarding without
unnecessary data transmission, the network coding is used by Chachulski et al.
In MORE, some predefined number of packets compose a batch as that in ExOR. All pack-
ets in the same batch will be encoded by linear network coding before sending out from the
source node. The source node will calculate a sorted Forwarder List by the routing metric of
250  Opportunistic Networks

expected transmission count (ETX) [18], which is also quite similar to ExOR. When a node
in the Forwarder List receives a packet of a batch, and if this packet is linear independent
from all received packets in the same batch, the packet should be forwarded. The exact time
to forward the packet depends on two factors: one factor is the position of the forwarder in
the Forwarder List, and the other factor depends on the 802.11 MAC. The difference from
ExOR is that the time used to wait for transmission can be shortened, whereas if the timer
expires and the medium is free, the linear encoded packet can be transmitted.

13.2.1.2 Other Variants of Opportunistic Forwarding

Besides the contributions on coordination protocol for opportunistic data forwarding, sev-
eral works make many other trials by changing the network types or metrics, or combining
with other technologies to help opportunistic data forwarding make decisions.

13.2.1.2.1 Network Types
A multi-hop wireless network is like a family which is composed by mesh networks, mobile
ad hoc networks, sensor networks, and vehicular networks. Each network can borrow the
idea of opportunistic data forwarding from mesh networks to promote theirs performance.
By following this way, Ma et al. [19] uses opportunistic forwarding in sensor networks to
enhance the probability that a data packet is forwarded successfully. Hence, more sensor
nodes can remain in the sleep mode for a longer time to save energy without influencing per-
formance. Vehicular networks can also explore the benefits of opportunistic data forwarding,
and usually they use the positioning service to get the topology of the vehicular networks and
select the best forwarder by the perspective of distance, like Leontiadis and Mascolo [20].

13.2.1.2.2 Metrics
In ExOR, ETX is used as the metric to evaluate which candidate is better. However, num-
bers of metrics have been proposed recently, such as distance, expected any-path transmis-
sions, and backlog. Distance is quite suitable for position-based opportunistic forwarding
because it is quite easily calculated by positioning service and has relatively good perfor-
mance in Leontiadis and Mascolo [20] and in Yang et al. [15]. Expected any-path transmis-
sions (EAX) is proposed in Zhong et al. [16], which are equal to the expected number of
transmissions required to deliver a packet to its destination under the perspective of oppor-
tunistic forwarding.

13.2.1.2.3 Network Coding-Based Opportunistic Forwarding

Network coding-based opportunistic data forwarding was mentioned in the previous section,
which usually uses the linear network coding method to encode the received or original pack-
ets in a series of random linear combinations. The exploration on this includes Chachulski
et al. [17], Radunović et al. [21], and Koutsonikolas et al. [22]. The coordination protocols
will not work as aggressively as those in ExOR. Usually, ACKs should be sent back when the
destination receives enough information to decode all packets in the batch, and the source
node will stop sending packets or go on sending the packets in the next batch when the ACK
for the current batch is received.

13.2.1.2.4 Position-Based Opportunistic Forwarding

Position based opportunistic forwarding is explored in Yang et al. [15]. The metric is the
distance from any forward candidates to the final destination, and the coordinate protocol
 Opportunistic Routing in Mobile Networks  251

is quite similar to that in ExOR. The node which is closer to the destination will access the
medium earlier to forward buffered packets.

13.2.1.3 Scenarios Suitable for Opportunistic Forwarding

In the previous sections, we have presented the advantages of opportunistic data forwarding
and the challenges in it. Both Shah et al. [23] and Kim et al. [24] make a comparison between
traditional IP forwarding and opportunistic data forwarding. They believe that the opportu-
nistic data forwarding is suitable for the networks whose mobility is not high, and the higher
the node density, the better performance that can be achieved. The reason to explain this is
that, if the mobility is high, the forwarder candidates cannot always keep in communication
range with each other, so a packet may be transmitted by two or more forwarders several
times. The duplication becomes the most important reason for the decreases in the benefits
we get from opportunistic data forwarding, and the performance gain by opportunistic for-
warding will be marginal or even negative because the coordination overhead always exists
in coordination protocols.

13.2.1.4 Performance Modeling for Opportunistic Forwarding

Many performance analyses and optimization works also contributed to the study of oppor-
tunistic data forwarding. These works run in two directions. One direction is the building of
models to analyze opportunistic data forwarding, and the other is using existing models to
enhance the performance in opportunistic data forwarding.

13.2.1.4.1 Performance Optimization
No matter what kind of data transmission protocols people choose to use, they want to
optimize the network’s performance. Moreover, if a boundary of performance can be proved,
it will be quite useful. Two performance optimization works have been done by Radunović
et al. [21] and Zeng et al. [25]. In opportunistic data forwarding, many forwarder candi-
dates can overhear a same packet, but a different forwarding decision will affect the perfor-
mance significantly. Hence, Radunović et al. uses an optimal flow-decision to maximize the
links’ utility of the whole network. A comprehensive model built by Radunović et al. gives
us the relationship between the links’ utility and flow-decision set. To maximize the utility,
Radunović et al. handle the optimization problem with two steps. First, they propose trans-
port credits, which are used to denote the quantity of packets that have been sent out. Second,
for each possible flow on each node, Radunović et al. studied the relationship between the
flow’s transport credits and a three-tuple composed by scheduling strategy, power control,
and transmit rate control. Hence, the transport credits for a particular node can be presented
by the three-tuple, so the summation of all nodes’ transport credits in the network can be
related with the three-tuple as well. When we take the three-tuple as the varying variable, the
optimized network utility is achieved by selecting the tuples that give us the maximum value
of transport credits.
Zeng et al. use “Transmitter Conflict Graph” rather than “Link Conflict Graph” to find
what the lower and upper bounds are of throughput in the opportunistic forwarding-based
multi-hop wireless networks. Three important concepts are introduced in Zeng et al. [25].
They are Concurrent Transmitter Sets (CTS), Conservative CTS (CCTS), and Greedy CTS
(GCTS). CTS is self-explained, and CCTS is one extreme case in CTS that all links associated
with any node in the CCTS can be used simultaneously. GCTS is another case in CTS that at
least one link of any node in the node set can be used concurrently. Hence, the lower bound
252  Opportunistic Networks

and the upper bound can be calculated in GCTS and CCTS separately. The lower bound
can be calculated in the protocol model with linear programming, and the upper bound can
be figured out in the physical model because of the fact that SNR will decrease when more
concurrent links are used.

13.2.1.4.2 Modeling from Markov Process and Game Theory

Cerdà-Alabern et al. [26] and Lu et al. [27] are two typical works based on the Markov
Process. The idea lies on the packet-received probability between every node pair in a
network that can be estimated, either by radio’s propagation model or by exchanging the
probe messages periodically. Moreover, in opportunistic forwarding, the node with the
smallest metric should forward the data first, so we can predict the performances statisti-
cally, such as hop count, transmission count, and even energy consumption. In Lu et al.
[27], the Markov Process is constructed with the states which are the combinations of dif-
ferent events, and the probability distribution of the ETX for a node to deliver a packet is
proposed. However, Cerda-Alabern et al. [26] use just the forwarder candidates to be the
states in Markov Process. The receive probability between every pair of nodes is estimated
by the radio propagation model. Therefore, the expectations of hop count and throughput
could be found.
Game Theory is used in Wu et al. [28], and the purpose of their work is to try to build
a mechanism by which the rewards and punishments are given to each node to make it
honestly announce its link quality and help other nodes forward packets. The reason to do
such work is that nodes in opportunistic forwarding may be selfish, and such a behavior
is unfair to other nodes and degrades the total performance. Wu et al. [28] use the Strict
Dominant Strategy Equilibrium to design a mechanism in which nodes will get the most
benefit if and only if they honestly report link quality and do their best to forward packets
for others.

13.2.2 Routing Algorithms Review for Opportunistic Data

Forwarding in MANETs
In this section, we will briefly review the existing routing protocols. The routing protocol is
important for our CORMAN because we want to utilize the Route-Prioritized Contention
(Section 13.2.1.1.2) approach to equip mobile ad hoc networks (MANETs) with the ability
of opportunistic data forwarding. This thesis aims to present CORMAN, a series of solutions
of opportunistic data forwarding in MANETs, so we will look at whether the existing rout-
ing protocols could support opportunistic data forwarding and how they would perform if
they could.

13.2.2.1 Timing Strategy in Routing Protocols

Basically, routing protocols in MANETs can be categorized using an array of criteria. The
most fundamental among these is the timing of routing information exchange. On one hand,
a protocol may require that nodes in the network should maintain valid routes to all destina-
tions all the time. In this case, the protocol is considered to be proactive, a.k.a. table driven.
Examples of proactive routing protocols include Destination-Sequenced Distance-Vector
(DSDV) [3] and Optimized Link State Routing (OLSR) [4]. On the other hand, if nodes in
the network do not always maintain routing information, when a node receives data from
the upper layer for a given destination, it must first find out how to reach the destination.
This approach is called reactive, a.k.a. on demand. Dynamic Source Routing (DSR) [5] and
Ad Hoc On-Demand Distance Vector (AODV) [6] fall in this category.
 Opportunistic Routing in Mobile Networks  253

13.2.2.2 Fundamental Algorithms – Link State and Distance Vector

These well-known routing schemes can also be categorized by their fundamental algorithms.
The most important algorithms in routing protocols are Distance Vector (DV) and Link State
(LS) algorithms. In LS, every node will share its best knowledge of links with other nodes in
the network, so nodes can reconstruct the topology of the entire network locally, e.g., OLSR.
In DV, a node only provides its neighbors with the cost to every given destination, so nodes
know the costs to a given destination via different neighbors as the next hop, e.g., DSDV and
AODV. DSR is an early source routing protocol in MANETs. In DSR, when a node has data
for a destination node, a route request is flooded to all nodes in the network. An intermediate
node that has received the request adds itself in the request and rebroadcasts it. When the
destination node receives the request, it transmits a route reply packet to send the discovered
route back to the source node via the route in request packet reversely, and the source node
will use the discovered route in data transmission.

13.2.2.3 Tree-Based Routing Protocols Derived from the Internet

As the scalability of the Internet also suffers from the overhead problem, many lightweight
routing protocols for the Internet have been proposed. The pioneer work by Garcia-Luna-
Aceves and Murthy [29] proposed a new routing algorithm called Loop-Free Path-Finding
Algorithm (LPA), based on the path-finding algorithm (PFA). To avoid loops, LPA proposed a
Feasibility Condition (FC). The FC can effectively avoid the temporary loop but may overkill
some possible valid paths in the network. Another piece of work, by Behrens and Garcia-
Luna-Aceves [30], proposed a new routing algorithm called Link Vector (LV). In LV, the basic
element in the update message contains the destination, the precursor to destination, the cost
between them, and a sequence number to avoid loops. LV is different from LS, because in LS,
an updater sends all the link information it knows by flooding; but in LSA, an updater only
sends the information of links which are preferred. So, the node in LV can construct only a
partial topology of the entire network. An algorithm proposed by Levchenko et al. [31] called
approximate link state (XL) conclusively summarizes the LV and PFA in a nutshell. It uses
soundness and completeness to define the correctness of the routing protocols, and it uses
stretch to evaluate the optimality. XL also uses the lazy update concept to further decrease
the overhead. Hence, in XL, a node sends an update message that contains only the changed
links which are on the only way that must be passed to a given destination, or the updated
link can improve the cost to a destination node by a given parameter.
Similar trails to reduce wireless routing overhead have been made during the develop-
ment listed previously. Murthy and Garcia-Luna-Aceves proposed the Wireless Routing
Protocol (WRP) [32], which utilizes the basic idea of PFA within wireless networks. Every
node in WRP has a tree structure for the network, and every time the node sends out only
the routing update message by differential update. However, WRP requires the receiver of
an update message to transmit the ACK. Such a requirement introduces more overhead and
consumes more channel resources, but it may not improve the performance dramatically.
Furthermore, WRP uses RouterDeadInterval to detect the loss of neighbors which may
adversely affect the response time of topology change, considering only one route is main-
tained in the tree structure. An extensional work based on WRP is Source Tree Adaptive
Routing (STAR) [33], which is proposed by Garcia-Luna-Aceves and Spohn. In STAR, every
node maintains a tree structure for the network and adopts a tree update strategy that is
neither proactive nor reactive. Instead, it uses a lazy approach, where an update message
will be transmitted only when the local tree structure is considered sufficiently inferior to
the original optimum.
254  Opportunistic Networks

13.2.2.4 Suitability of Existing Routing Protocols for Opportunistic Data Forwarding

in MANETs
In fact, none of the previous listed protocols can ideally support opportunistic data forward-
ing in MANETs. In particular, AODV [6], DSDV [3], and other DV-based routing algorithms
were not designed for source routing and hence are not suitable for opportunistic data for-
warding. The reason is that every node in these protocols knows only the next hop to reach a
given destination node, not the complete path. OLSR [4] and other LS-based routing proto-
cols could support source routing, but their overhead is still fairly high for the load-sensitive
MANETs. DSR together with its derivations, such as [34], [35], and [36], are not suitable
because they have a long bootstrap delay and are therefore not efficacious for frequent data
exchange. Furthermore, the reactive routing protocols will inject too much route request
packets in the mobile networks, especially when there are a large number of data sources.
Moreover, the route reply message may be lost since it is sent based on IP forwarding via
recorded route reversely, so reactive routing schemes suffer from even more unpredictable
delay in data transmission. The WRP [32] and STAR [33] – the early attempts to port the
routing capabilities in linking state routing protocols to MANETs – are built on the frame-
work of PFA for each node to use a tree for loop-free routing. Although WRP is an innovative
exploration in the research on MANETs and STAR uses a “lazy” routing strategy to reduce
the topology update burden, they have a very high operation overhead due to the routing
procedures being triggered by the event of topology changes, which introduce a great deal
of information exchanged and stored by the nodes. Our intention was to include WRP in
our experimental comparison later in this chapter, and we have implemented WRP in NS-2.
Unfortunately, our preliminary tests indicate that its communication overhead is at least an
order of magnitude higher than the other main-stream protocols.
In a nutshell, designing a new lightweight proactive source routing protocol is a very
important component to develop opportunistic data forwarding in MANETs.

13.2.3 Motivation and Framework

When we review the related work, we can see that a systematic solution to equip MANETs
requires the ability of opportunistic data forwarding. Hence, in our research, we propose
CORMAN as a network layer solution to implement opportunistic data forwarding in
MANETs. Its node coordination mechanism is largely inline with that of ExOR, and it is
an extension to ExOR in order to accommodate node mobility. Here, we first highlight our
objectives and challenges in order to achieve them. Later in this section, we provide a general
description of CORMAN. The details of the major components of CORMAN are presented
in Sections 13.3 (proactive source routing), 13.4 (large-scale live update), and 13.5 (small-
scale retransmission).

13.2.3.1 Objectives and Challenges

CORMAN has two objectives: (1) it broadens the applicability of ExOR to mobile multi-hop
wireless networks without relying on external information sources, such as node positions;
and (2) it incurs a smaller overhead than ExOR by including shorter forwarder lists in data
packets.
The following challenges are thus immanent.

1. Overhead in route calculation: CORMAN relies on the assumption that every source
node has complete knowledge of how to forward data packets to any node in the net-
work at any time. This calls for a proactive source routing protocol. Link state routing,
 Opportunistic Routing in Mobile Networks  255

such as OLSR, would meet our needs, but it is fairly expensive in terms of communica-
tion costs. Therefore, we need a lightweight solution to reduce the overhead in route
calculation.
. Forwarder list adaptation: When the forward list is constructed and installed in a data
2
packet, the source node has updated knowledge of the network structure within its
proximity, but its knowledge about further areas of the network can be obsolete due
to node mobility. This becomes worse as the data packet is forwarded towards the des-
tination node. To address this issue, intermediate nodes should be able to update the
forwarder list adaptively with their new knowledge when forwarding data packets.
3. Robustness against link quality variation: When used in a dynamic environment, a
mobile ad hoc network must inevitably face the drastic link quality fluctuation. A short
forwarder list carried by data packets implies that they tend to take long and possibly
weak links. This could be problematic for opportunistic data transfer, since the list may
not contain enough redundancy in selecting intermediate nodes. This should be over-
come with little additional overhead.

13.2.3.2 CORMAN Fundamentals
CORMAN forwards data in a similar batch-operated fashion as ExOR. A flow of data pack-
ets are divided into batches. All packets in the same batch carry the same forwarder list when
they leave the source node. To support CORMAN, we have an underlying routing protocol,
Proactive Source Routing (PSR), which provides each node with the complete routing infor-
mation to all other nodes in the network. Thus, the forwarder list contains the identities of
the nodes on the path from the source node to the destination. As packets progress in the
network, the nodes listed as forwarders can modify the forwarder list if the network topology
has observed any changes. This is referred to as large-scale live update in our work. In addi-
tion, we also allow some other nodes that are not listed as forwarders retransmit data if this
turns out to be helpful, referred to as small-scale retransmission. Note that CORMAN is a
complete network layer solution and can be built upon off-the-shelf IEEE 802.11 networking
commodities without any modification.
Therefore, the design of CORMAN has the following three modules. Each module answers
to one of the challenges stated previously.

1. Proactive source routing: PSR runs in the background so that nodes periodically
exchange network structure information. It converges after the number of iterations
equal to the network diameter. At this point, each node has a spanning tree of the net-
work, indicating the shortest paths to all other nodes. The amount of information broad-
cast by each node in an iteration is O(n), where n is the number of nodes in the network.
Such an overhead is the same as distance vector algorithms and is much smaller than
link state. Technically, PSR can be used without CORMAN to support conventional IP
forwarding.
2. Large-scale live update: When data packets are received by and stored at a forwarding
node, the node may have a different view of how to forward them to the destination
from the forwarder list carried by the packets. Since this node is closer to the destina-
tion than the source node, such discrepancy usually means that the forwarding node
has more updated routing information. In this case, the forwarding node updates the
part of the forwarder list in the packets from this point on towards the destination
according to its own knowledge. When the packets with this updated forwarder list are
broadcast by the forwarder, the update of network topology change propagates back
to the upstream neighbor. The neighbor incorporates the change to the packets in its
256  Opportunistic Networks

cache. When these cached packets are broadcast later, the update is further propagated
towards the source node. Such an update procedure is significantly faster than the rate
at which a proactive routing protocol disseminates routing information.
. Small-scale retransmission: A short forwarder list forces packets to be forwarded over
3
long and possibly weak links. To increase the reliability of data forwarding between
two listed forwarders, CORMAN allows nodes that are not on the forwarder list but
are situated between these two listed forwarders to retransmit data packets if the down-
stream forwarder has not received these packets successfully. Since there may be mul-
tiple such nodes between a given pair of listed forwarders, CORMAN coordinates
retransmission attempts among them extremely efficiently.

13.3 PROACTIVE SOURCE ROUTING – PSR

Before describing the details of PSR, we will first review some graph theoretic terms used
here. Let us model the network as an undirected graph G = (V, E), where V is the set of nodes
(or vertices) in the network and E is the set of wireless links (or edges). Two nodes u and v
are connected by an edge e = (u, v) ∈ E if they are close to each other and can communicate
directly with a given reliability. Given a node v, we use N(v) to denote its open neighborhood,
i.e., {u ∈ V| (u, v) ∈ E}. Similarly, we use N[v] to denote its closed neighborhood, i.e., N(v) ∪
{v}. In general, for a node v, we use Nl(v) (l ≥ 1)) to denote the distance-l open neighborhood
of v, i.e., the set of nodes that are exactly l hops away from v. Similarly, Nl[v] (l ≥ 1) denotes
the distance-l closed neighborhood of v, i.e., the set of nodes that are within l hops of v. As
special cases, N1[v] = N[v], N1(v) = N(v), N0[v] is v itself, and N0(v) = ∅. Also as a convention
in graph theory, for any S ⊆ V, we use 〈S〉 to denote the subgraph induced by S, i.e., 〈S〉 =
(S, E′), where E′ is the set of edges where each element has both endpoints in S. The readers
should refer to West [37] for other graph theoretic notions and other details.
The research findings from CORMAN entirely have been published to the IEEE Journal
on Selected Areas in Communications [38]. The research findings from PSR and its early
refinement have been published to IEEE GLOBECOM ’11 [39] and IEEE Communications
Letters [40]. PSR’s last improvement was submitted to IEEE INFOCOM ’12 [41]

13.3.1 Design of PSR
Essentially, PSR provides every node with a Breadth First Spanning Tree (BFST) of the entire
network rooted at itself. To do that, nodes periodically broadcast the tree structure to its best
knowledge in each iteration. Based on the information collected from neighbors during the
most recent iteration, a node can expand and refresh its knowledge about the network topol-
ogy by constructing a deeper and more recent BFST. This knowledge will be distributed to
its neighbors in the next round of operation (Section 13.3.1.1). On the other hand, when a
neighbor is deemed lost, a procedure is triggered to remove its relevant information from the
topology repository maintained by the detecting node (Section 13.3.1.2). Intuitively, PSR has
about the same communication overhead as distance-vector-based protocols. We go the extra
mile to reduce the communication overhead incurred by PSR’s routing agent. Details about
such overhead reduction will be discussed in Section 13.3.1.3.

13.3.1.1 Route Update
Due to its proactive nature, the update operation of PSR is iterative and distributed among
all nodes in the network. At the beginning, a node v is aware only of the existence of itself,
 Opportunistic Routing in Mobile Networks  257

so there is only a single node in its BFST, which is the root node v. By exchanging the BFSTs
with the neighbors, it is able to construct a BFST within N[v], i.e., the star graph centered at
v, denoted by Sv.
In each subsequent iteration, nodes exchange their spanning trees with their neighbors.
From the perspective of node v, towards the end of each operation interval, it has received
a set of routing messages from its neighbors packaging the BFSTs. Note that, in fact, more
nodes may be situated within the transmission range of v, but their periodic updates were
not received by v due to, say, bad channel conditions. After all, the definition of a neighbor
in MANETs is a fickle one. (We have more details on how we handle lost neighbors subse-
quently.) Node v incorporates the most recent information from each neighbor to update its
own BFST. It then broadcasts this tree to its neighbors at the end of the period. Formally, v
has received the BFSTs from some of its neighbors. Including those from whom v has received
updates in recent previous iterations, node v has a BFST, denoted Tu, cached for each neigh-
bor u ∈ N(v). Node v constructs a union graph

Gv  Sv 
 T  v .
uN  v 
u (13.1)

Here, we use T − x to denote the operation of removing the subtree of T rooted at node
x. As special cases, T − x = T if x is not in T, and T − x = ∅ if x is the root of T. Then, node
v calculates a BFST of Gv, denoted Tv, and places Tv in a routing packet to broadcast to its
neighbors.
In fact, in our implementation, the aforementioned update of the BFST happens multiple
times within a single update interval so that a node can incorporate new route information
to its knowledge base more quickly. To the extreme, Tv is modified every time a new tree is
received from a neighbor. Apparently, this is a trade-off between the routing agent’s adap-
tivity to network changes and computational cost. Here, we choose routing adaptivity as a
higher priority assuming that the nodes are becoming increasingly powerful in packet pro-
cessing. Nevertheless, this does not increase the communication overhead at all because one
routing message is always sent per update interval.
Assume that the network diameter, i.e., the maximum pairwise distance, is D hops. After
D iterations of operation, each node in the network has constructed a BFST of the entire
network rooted at itself. This information can be used for any source routing protocol. The
amount of information that each node broadcasts in an iteration is bounded by O(| V| ) and
the algorithm converges in at most D iterations.

13.3.1.2 Neighborhood Trimming
The periodically broadcast routing messages in PSR also double as “Hello” messages
for a node to identify which other nodes are its neighbors. When a neighbor is deemed
lost, its contribution to the network connectivity should be removed, called “neighbor
trimming”. Considering node v, the neighbor trimming procedure is triggered at v about
neighbor u when

• no routing update or data packet has been received from this neighbor for a given
period of time, or
• a data transmission to node u has failed as reported by the link layer.
258  Opportunistic Networks

Node v responds by

1. first, updating N(v) with N(v) − {u};

2. next, constructing the union graph with the information of u removed

Gv  Sv 
 T
wN  v 
w  v,

(13.2)

3. then, computing the BFST Tv.

Notice that Tv thus calculated is not broadcast immediately to avoid excessive messaging.
With this updated BFST at v, it is able to avoid sending data packets via lost neighbors. Thus,
multiple such neighbor trimming procedures may be triggered within one period.

13.3.1.3 Streamlined Differential Update

In addition to dubbing route updates as hello messages in PSR, we interleave the “full dump”
routing messages as stated previously with “differential updates”. The basic idea is to send
the full update messages less frequently than shorter messages, containing the difference
between the current and previous knowledge of a node’s routing module. Both the benefit of
such an approach and how to balance between these two types of messages have been stud-
ied extensively in earlier proactive routing protocols. In this work, we further streamline the
routing update in two new avenues. First, we use a compact tree representation in full dump
and differential update messages to halve the size of these messages. Second, every node
attempts to maintain a “stable” BFST as the network changes, so that the differential update
messages are even shorter.

• Compact tree representation: For the full dump messages, our goal is to broadcast the
BFST information stored at a node to its neighbors in a short packet. To do that, we
first convert the general rooted tree into a binary tree of the same size, say s nodes. Then
we serialize the binary tree using a bit sequence of 34 × s bits, where the IPv4 (Internet
Protocol version 4) is assumed. Specifically, we scan the binary tree layer by layer. When
processing a node, we first include its IP address in the sequence. In addition, we append
two more bits to indicate if it has the left and/or right child. For example, the binary
tree in Figure 13.4 is represented as A10B11C11D10E00F00G11H00I00. As such, the
size of the update message is a bit over half compared to the traditional approach,
where the message contains a discrete set of edges.

C D

E F G

Figure 13.4  Binary tree.

 Opportunistic Routing in Mobile Networks  259

The difference between two BFSTs can be represented by the set of nodes who have
changed parents, which are essentially a set of edges connecting to the new parents. We
observe that these edges are often clustered in groups. That is, many of them form a
sizable tree subgraph of the network. Similar to the case of full dump, rather than using
a set of loose edges, we use a tree to package the edges connected to each other. As a
result, a differential update message usually contains a few small trees, and its size is
noticeably shorter.
• Stable BFST: The size of a differential update is determined by how many edges it
includes. Since there can be a large number of BFSTs rooted at a given node of the
same graph, we need to alter the BFST maintained by a node as little as possible when
changes are detected. To do that, we modify the computation described earlier in this
section, such that a small portion of the tree needs to change either when a neighbor is
lost or when it reports a new tree.
Consider node v and its BFST Tv. When it receives an update from neighbor u,
denoted by Tu, it first removes the subtree of Tv rooted at u. Then it incorporates the
edges of Tu for a new BFST. Note that the BFST of (Tv − u) ∪ Tu may not contain all
necessary edges for v to reach every other node. Therefore, we still need to construct
the union graph

Tv  u    T
wN  v 
w  v,

(13.3)

before calculating its BFST. To minimize the alteration of the tree, we add one edge
of (Tw − v) to (Tv − u) at a time. During this process, when there is a tie, we always try
to add edges that were originally removed from Tv.
When node v thinks that a neighbor u is lost, it deletes the edge (u, v) but still utilizes
the network structure information contributed by u earlier. That is, even if it has moved
away from v, node u may still be within the range of one of v’s neighbors. As such, Tv
should be updated to a BFST of

Tv  u   Tu  v    T
wN  v 
w  v .

(13.4)

Note that, since N(v) no longer contains u, we need to explicitly put it back into the
equation. Similarly in this case, edges of (Tu − v) ∪ ∪w ∈ N(v)(Tw − v) are added to (Tv − u)
one at a time, with those just removed because of u taking priority.

13.3.2 Implementation
As the implementation contains many algorithms, here we can only introduce some impor-
tant ones. Before looking at the algorithms, we summarize the notations we used for these
algorithms in Table 13.1.

13.3.2.1 Routing and Neighborhood Update Algorithm

As the operations in routing update and neighborhood update are similar, we take the
routing update as an example to introduce their implantation. Equation 13.3 theoretically
enables us to implement the routing update algorithm. However, direct implementation
needs a high algorithm complexity, and, according to the procedure in Equation 13.3, every
time a node v receives an update from neighbor uk, we have to check every node in subtrees
Tw − v (w ∈ N(v)). That is a time-consuming job in both reality and simulation. In fact,
260  Opportunistic Networks

Table 13.1  Notations in algorithms.

Notation Explanation

stack ⇑ [v] Pop the top of stack and store to v, v is optional

stack ⇓  v Push v on the top of stack
queue ⇑ [v] Get an element from FIFO queue and keep in optional v
queue ⇓  v Put v into queue in FIFO manner
R(T) Get the root node of tree T
Bl/r(n) The left/right brother node of n
LT(n) Look up node with same ID of n from tree T
Gs/b/l(n) Create a spanning/binary tree node or linear element of n
AT(np, nc) Add a new node nc as a child of np in T
UT(np, nc) Update np as the new parent of nc in tree T
HT(n) Hops from node n to its root node in tree T
Cx(n) The xth child of node n in spanning tree
Cl/r(n) The left/right child of node n in binary tree
P(n) The parent of node n
cl(e) The left tag of a linear element e
cr(e) The right tag of a linear element e
a∪e append an element e at the end of an array a

from Section 13.3.1.1 we can see that the shorter path is always selected, and if two paths
with same hops are found in the union graph, we always keep the original one. Hence, in
our implementation, we look up every node in trees Tw − v (w ∈ N(v)) from the BFST being
constructed. If a duplicated node is found and the new discovered path is shorter, the parent
of the node should be changed. Therefore, the incorporating work for every subtree Tw − v
(w ∈ N(v)) can be finished by Algorithm 13.1.
The routing update for neighbor uk would be finished if we run Algorithm 13.1 for all
subtrees in the last parameter of Equation 13.3. Hence the implementation of neighborhood
update would be finished by the same way, and the only difference is the initial Tv.

13.3.2.2 Algorithms for Transformation of Tree Structures

In general, each node maintains the topology by BFST as we introduced before, which is quite
suitable to provide a route to a destination node. When a routing update packet is needed,
either the entire spanning tree in a full dump update or the forest in a differential update
should be presented in linear form. In fact, a single tree can be taken as a forest with one
member. As we talked before, we first convert a spanning forest to a binary forest, and then
convert the binary forest into its linear form. When a node receives the update, the receiver
should restore the binary form of a linear expression and then transform the binary forest to
a spanning forest. Depending on which kind of update is used, we can choose how to deal
with the spanning forest. If the update is a full dump one, there must be only one tree in the
spanning forest, and we direct replace the original BFST cached for such a neighbor by the
new one. Otherwise, we should update the original BFST according to the spanning forest
as introduced in Section 13.3.2.3. In this part we talk only about the algorithms related with
the transformation of tree structures. In a nutshell, we have five algorithms to help us finish
all these work:
 Opportunistic Routing in Mobile Networks  261

ALGORITHM 13.1  INCORPORATE  ui  v  ui  N  v   TO  v

stack ⇓ R  ui  v 
loop
  stack ⇑ nuci
  if nuci is null then
  if stack is not null then
   stack ⇑ nuci ; stack ⇓ Br nuci 
  else
   return;
  end if
  else
  nvc ⇐ Lv  nuci 
  if stack is null then
   nvp ⇐ R(T v)
  else
  stack ⇑ nupi ; nvp ⇐ Lv  nupi ; stack ⇓ nupi
  end if
  if nvc is null then
nvc ⇐ Gs  nuci  ; A v  nvp ,nvc 
  else if H v  nvc   H  ui  v  nuci  + 1 then
   U v  nvp ,nvc 
  end if
  stack ⇓ nuci
  if nuci is leaf node then
   stack ⇓ null
  else
   stack ⇓ C0 nuci 
  end if
  end if
end loop

• Algorithm to convert spanning tree to binary tree – Algorithm 13.2.

• Algorithm to convert binary tree to spanning tree – Algorithm 13.3.
• Algorithm to convert binary tree to linear tree – Algorithm 13.4.
• Algorithm to convert linear tree to binary tree – Algorithm 13.5.
• The last one is the ancillary algorithm to separate the linear-forest to a set of

linear-trees – Algorithm 13.6.

13.3.2.3 Implementation of Reconstruction in Differential Update

In this part, we will cover the algorithm used on the receiver side to reconstruct the neigh-
bor’s BFST in a differential update. The implementation of setting up a differential update
can be more easily introduced after talking about Algorithms 13.1, 13.4, 13.5, and 13.6.
262  Opportunistic Networks

By operating all four of these algorithms, the linear-forest in a differential update can be
separated into a set of linear-trees, and each linear-tree can be converted to a binary tree
and further to a spanning tree. As we specified before, every edge in a differential spanning
tree indicates a new parent in a neighbor’s BFST; hence, the differential update can be easily
handled by incorporating all spanning trees in the local cached one. The incorporation pro-
cess is similar with Algorithm 13.1; however, there are two differences.

1. We do not check the hop count of path anymore. Instead, we replace all edges with the
new ones indicated by the differential spanning tree.
2. A spanning tree with root ID 255.255.255.255 is a spatial case in the differential update.
All nodes in such spanning tree should be trimmed from the BFST of the neighbor.

ALGORITHM 13.2  CONVERT SPANNING TREE TV TO BINARY TREE TV′

Initialize Tv′ with node v as root
stack ⇓ R(Tv)
wvc′ ⇐ R(Tv′)
loop
  stack ⇑ nvc
  if nvc is null then
  if stack is not null then
   stack ⇑ nvc
   for i = 1 to Nc(nvc ) do
    wvc  P  wvc 
   end for
   nvc  B r  nvc 
   if nvc is not null then
    nvc  Gb  nvc  ; P  nvc   wvc ; C r  wvc   nvc ; wvc  nvc
   end if
   stack ⇓ nvc
  else
   return;
  end if
  else if nvc is leaf node then
  stack ⇓ null
  else
  nvc  C0  nvc  ; nvc  Gb  nvc  ; P  nvc   wvc ; Cl  wvc   nvc ; wvc  nvc ;
  stack ⇓ nvc
  end if
end loop
 Opportunistic Routing in Mobile Networks  263

ALGORITHM 13.3  CONVERT BINARY TREE TV TO SPANNING TREE TV′

Initialize Tv′ with node v as root
Initialize tag ⇐ 0 (tag ∈ {0, 1, 2})
stack ⇓ R(Tv)
wv  R   v 
c

if wvc′ is leaf node then

 return;
else
  loop
  stack ⇑ nvc
  if nvc is null then
   if tag = 0 then
    stack ⇑; tag ⇐ 1
   else if tag = 1 then
    stack ⇑; tag ⇐ 2
   else if tag = 2 then
    return;
   end if
  else
   if tag = 0 then
    if Cl  nvc  is not null then
 
     stack ⇓ nvc ; stack ⇓ Cl  nvc  ; nvc  Gs C1  nvc  P  nvc   wvc;
     C Nc wc  wvc   nvc ; wvc  nvc ; tag  0
 v 
    else
     stack ⇓ null; tag ⇐ 0
    end if
   else if tag = 1 then
    if C r  nvc  is not null then
 
     stack ⇓ nvc ; stack ⇓ C r  nvc ; nvc  Gs C r  nvc  ; P  nvc   P  wvc  ;
 
     C N P wc P  wvc   nvc ; wvc  nvc ; tag  0
c   v  
    else
     stack ⇓ null; tag ⇐ 1
    end if
   else if tag = 2 then

    if nvc  C1 P  nvc  then 
     tag ⇐ 1; wvc  B1  wvc 
    else
     tag ⇐ 2; wvc  P  wvc 
    end if
    stack ⇑;
    if wvc′ is null then
     return;

(Continued)
264  Opportunistic Networks

    end if
   end if
  end if
  end loop
end if

ALGORITHM 13.4  CONVERT BINARY TREE Tv TO A LINEAR-ELEMENT

ARRAY ae
Initialize ae with null
queue ⇓ R(Tv)
while queue is not null do
  queue ⇑ nv
  eae ⇐ G1(nv)
  if Cl(nv) is null then
  cl(eae) ⇐ ‘ 0’
  else
  cl(eae) ⇐ ‘1’; queue ⇓ Cl(nv)
end if
  if Cr(nv) is null then
  cr(eae) ⇐ ‘0’
  else
  cr(eae) ⇐ ‘1’; queue ⇓ Cr(nv)
  end if
  ae ⇐ (ae ∪eae)
end while

ALGORITHM 13.5  CONVERT LINEAR-ELEMENT ARRAY ae TO BINARY

TREE Tv
Require: ae is not null
 Initialize index = 0; m1 = –1; m2 = 0
  Initialize binary-node array an with null
  nv ⇐ Gs(ae[index + +])
 R(Tv) ⇐ nv
  an[0] ⇐ R(Tv)
  while m1 ≠ m2 do
  cnt ⇐ 0
  for i = 1 to m2 – m1 do
   nvp ⇐ an[m1 + i]
   if Cl(nvp) is ‘1’ then
 Opportunistic Routing in Mobile Networks  265

    nvc ⇐ Gs(ae[index + +])

    Cl  nvp   nvc ; P  nvc   nvp ; an  an  nvc ; cnt  
   end if
   if C r  nvp  is ‘1’ then
    nvc ⇐ Gs(ae[index + +])
    C r  nvp   nvc ; P  nvc   nvp ; an  an  nvc ; cnt  
   end if
   m1 = m2
   m2 = m2 + cnt
  end for
end while

ALGORITHM 13.6  CALCULATE THE ARRAY OF INDEXES TO SEPARATE ae

INTO LINEAR-TREES
Require: ae is not null
 Initialize milestone_array with null; milestone_cnt = 0; storage = 0
  milestone_array[milestone_cnt + +] = 0
  for i = 0 to |ae| – 1 do
  if cl(ae[i]) then
   storage + +;
  end if
  if cr(ae[i]) then
   storage + +;
  end if
  if storage = i then
   milestone_array[milestone_cnt + +] = i + 1; storage + +
  end if
  end for

13.4 LARGE-SCALE ROUTING UPDATE AND SMALL-SCALE

RETRANSMISSION

13.4.1 Large-Scale Live Update

When a batch of packets are forwarded along the route towards the destination node, if an
intermediate node is aware of a new route to the destination, it is able to use this new route
to forward the packets that it has already received. There are a few implications of this.
First, this new route will also be used to forward the subsequent packets of the same batch.
Second, when packets are forwarded along the new route, such an updated forwarder list
replaces the old list in the packets. As a result, the upstream nodes can be notified of the new
route, and this information can propagate back to the source node quickly. Details of data
forwarding and list update are described in this section with the help of Figure 13.5. In the
figure, the source node v1 has a flow of data packets for destination node v10. According to
266  Opportunistic Networks

v6 v7 v8

v1 v2 v3 v4 v5 v9 v10

v6 v7

Figure 13.5  Route update.

its own routing module, v1 decides that the best route to v10 is v1v2v3v4v5v6v7v8v9v10; hence,
the forwarder list.
At a given time point during the data transfer of a batch, there is a node on the forwarder
list that has the highest priority and has received any packet of the batch. We call such a
node the frontier of the batch. At the beginning, the frontier is the source node. When the
destination has received at least one packet of a batch, it has become the frontier of the batch.
Recall that a fragment (Section 13.2.1.1.2) is a subset of packets in the current batch which
are sent together from a given forwarder. Here, the frontier has cached its first fragment of
packets. Suppose at this point, the frontier in Figure 13.5 is v3. When it is about to forward
this fragment, if its routing module indicates there is a new route to the destination, e.g.,
v3v4v5v6′ v7′ v9v10 , it replaces the segment of the original forwarder list from itself to the desti-
nation (i.e., v3v4v5v6v7v8v9v10) with this new route. That is, the forwarder list carried by these
data packets is now v1v2v3v4v5v6′ v7′ v9v10. When the packets of the fragment are forwarded,
they will follow the new route. In addition, upstream nodes can overhear these packets, and
thus their new forwarder list. These nodes can update their own routing information and will
incorporate such information when forwarding their fragments. This backtrack continues
until the source is aware of the latest route information.
We would like to bring the following notes to the readers’ attention:

1. When the network diameter is large and nodes are moving fast, the routing information
can be obsolete by the time it has propagated to a remote node. That is, a node’s knowl-
edge about the network topology becomes less accurate when the destination node is
located farther away. Thus, the forwarder list composed by the source node needs to be
adjusted as packets are forwarded towards the destination, where intermediate nodes
closer to the destination could have better routing information. This is achieved effec-
tively by allowing the frontier node to modify the forwarder list carried by the fragment
of packets. As a result, CORMAN has a fairly good tolerance of route inaccuracy for
any source node to start with.
2. When a frontier node updates an forwarder list, only the segment of the list between
the frontier and destination is replaced, while the rest of the list (i.e., nodes that the
fragment has gone through) remains intact. The reason for this design decision is that
these upstream nodes should not be disturbed by the new route, so that the scheduling
coordination among them is consistent.
3. We allow a frontier node to update a packet’s forwarder list only according to its
routing module. A node that is no longer a frontier should incorporate only the for-
warder list that it overhears from downstream nodes. The purpose is to avoid unnec-
essary updates of route information, as the time needed to transfer a batch of data
packets is very short. During this time, usually little has changed about the network
topology, and nodes may not even have exchanged the routing information for the
next periodic interval.
 Opportunistic Routing in Mobile Networks  267

4. Consider a particular intermediate node on the forwarder list. As the frontier moves
from the source to the destination, the forwarder list may be refreshed multiple times
by different frontiers. Thus, this node may experience one update about its route to the
destination every time a frontier decides to modify the list.

All of these are achieved rapidly and with no extra communication overhead compared to
ExOR.

13.4.2 Small-Scale Retransmission
As we mentioned before, implementing opportunistic date transfer over the least hops path
can thoroughly explore the broadcast nature in wireless transmission; however, the penalty
of using such a path to transmit packets is that the connectivity between every pair of for-
warders may be vulnerable. We propose a solution to this problem, and the basic idea is that
the nodes not contained in the forwarder list [9] can run a distributed small-scale retransmit-
ter selection algorithm to verify whether it should participate in the packet forwarding. Such
a small-scale retransmitter should have the best position between two listed forwarders (the
forwarders which are contained in the forwarder list) compared with other nodes. In general,
our small-scale retransmitter selection algorithm has three important features. First, it is a
distributed algorithm. In particular, if there are many nodes between two listed forwarders,
after every node running the small-scale retransmitter selection algorithm, the best one will
know it is the best and participate in the data forwarding without sending any announce-
ment. Meanwhile, other nodes will directly refuse to be the small-scale retransmitter with-
out any notification, either. Second, the small-scale retransmitter selection algorithm is used
when the packets in a batch are received by the nodes which are not the forwarders in the
forwarder list, so the small-scale retransmitter can be selected on time without the problem
of inaccuracy. Third, all information required by the node to run, such as the algorithm, has
already been collected through exchange of routing packets, so no additional information
needs to be injected into the network.
In this section, we will first deeply analyze the reason why the small-scale retransmitter
is required in CORMAN and then specify the details in our solution. The research findings
from small-scale retransmission have been submitted to IEEE ICC ’12 [42]

13.4.2.1 Considerations of Small-Scale Retransmission

Previously, we proposed updating the forwarder list in a piggyback way when a data packet
is being forwarded towards the destination. However, it is not helpful to enforce the link con-
nectivity, because updating the forwarder list is helpful only when another node replaces the
one moving away; but if no node replaces the position, the forwarder list will still be broken,
as shown in Figure 13.6. Node Y moves from the position Y′ to the current position, with
the direction shown by the dashed arrow. Even though the wireless link is quite vulnerable, a
transient link from Y to X may exist and over such a link, node X sends downstream nodes

z
y
a
y' a,b
Dest X C B A a,b,c Src

Figure 13.6  Situation which cannot be solved by updating the forwarder list.
268  Opportunistic Networks

the information that the precursor to go to node Y is X. However, the vulnerable link is
unreliable or broken when the source node wants to use the forwarder list “Dest, Y, X, C, B,
A, Src”, even by opportunistic data forwarding. To deal with this problem, we propose that
node Z, which is a neighbor for both nodes X and Y, can be the small-scale retransmitter to
offer a help when it receives packets from upstream nodes. That is because if node Z receives
a packet with the forwarder list “Dest, Y, X, C, B, A, SrC”, Z can figure out that Y and Xare
the designated forwarders in the forwarder list, and meanwhile is itself a neighbor for both
of them. Hence, node Z could participate in forwarding the data packet. In particularly,
node Z also waits for a period of time before its forwarding. Such a time period should be
longer than the waiting time of node Y but shorter than the waiting time of node X. Or, if Z
can overhear a fragment of packets being transmitted from node Y, it can predict the exact
time when Y will finish its transmission and begin directly forwarding the rest of the useful
packet. However, before we discuss the details about the small-scale retransmitter selection
algorithm, it is necessary to introduce the changes of data structure kept on each node.

13.4.2.2 Design of Small-Scale Retransmission

In general, a Neighbor Table in CORMAN maintains the records of two hop topology
of the current node and the link quality from its neighbors to its neighbors’ neighbors. In
CORMAN, we use the Received Signal Strength Indicator (RSSI) to evaluate the quality of
links, because according to the work done by Charles Reis et al. [43] and Mei-Hsuan Lu et al.
[44], RSSI can be reported by almost all commodity wireless cards, and a greater RSSI value
usually indicates a better wireless channel quality.
In particular, RSSI Table is a sub-table maintained in every Neighbor Table entry which
records the RSSI values from that neighbor to that neighbor’s neighbors, and the entries in
the RSSI Table have the content shown in Figure 13.7. The Neighbor’s Neighbor ID keeps the
address of the nodes which are in fact two hop neighbors via current neighbor in Neighbor
Table; the RSSI records the RSSI value from the current neighbor to such two hop neighbors;
the Expire Time holds the time when the entry should be removed; and the next helps us find
the next entry in the same RSSI Table.
The operations on the Neighbor Table are presented as follows. When a routing update
packet needs to be transmitted out from one node periodically, it not only contains the rout-
ing information but also collects the RSSI values from all its local Neighbor Table entries.
All tuples (Neighbor ID, RSSI) are added into the routing packet as the Neighbor RSSI List.
After the node sends out such a packet, all its neighbors update the RSSI Tables kept by the
corresponding Neighbor Table entries.
To make the following discussion easier, we give a brief topology in Figure 13.8. In this
topology, we assume the wireless link from node X to node Y is a transient link, and such a
vulnerable link may successfully deliver a routing packet from X to Y or from Y to X. Hence,
the link between X and Y may be used as a part of the forwarder list by a node far away. In
this topology we can see the neighbors of node X are A, B, C, and Y, and node Y’s neighbors
are A, B, D and X. Therefore, nodes X and Y have the same two neighbors A and B. We use
R(M, N) to denote the RSSI value detected by node N from node M’s transmission, and to
make a intuitional discussion, the Neighbor Tables and their RSSI Tables on both node A and
B are briefly presented in Figure 13.9.

Neighbor's Neighbor ID RSSI Expire Time next

Figure 13.7  Entry of RSSI Table in Neighbor Table entry.

 Opportunistic Routing in Mobile Networks  269

p
D
A
X y

C
B
Q

Figure 13.8  Topology example.

X R(X, A) Establish Time RSSI table next X R(X, B) Establish Time RSSI table next

A R(A, X) Expire Time next A R(A, X) Expire Time next

B R(B, X) Expire Time next B R(B, X) Expire Time next
C R(C, X) Expire Time next C R(C, X) Expire Time next
Y R(Y, X) Expire Time next Y R(Y, X) Expire Time next

Y R(Y, A) Establish Time RSSI table next Y R(Y, B) Establish Time RSSI table next

A R(A, Y) Expire Time next A R(A, Y) Expire Time next

B R(B, Y) Expire Time next B R(B, Y) Expire Time next
D R(D, Y) Expire Time next D R(D, Y) Expire Time next
X R(X, Y) Expire Time next X R(X, Y) Expire Time next

Neighbor table on node A Neighbor table on node B

Figure 13.9  Neighbor table on node A and node B.

Without loss of generality, in Figure 13.8 we assume node A and node B overhear one or
more packets in a same batch, and node X and node Y are contained in the forwarder list in
these packets. In traditional opportunistic data forwarding, only the nodes in the forwarder
list can participate in the packet forwarding process, and the probability that the packet was
transmitted between X and Y may be quite low. That is because the least hop path may con-
tain unreliable and transient links in the forwarder list, and such a situation becomes even
worse in mobile cases. Such a problem can be solved by choosing one node from A and B
as the small-scale retransmitter, which is selected by small-scale retransmitter selection algo-
rithm as we present next.

13.4.2.3 Algorithm and Scoring Function

In particular, nodes A and B will decode the forwarder list from the overheard packet, if they
are not contained in the forwarder list (which is true in our assumption; otherwise, the node
will participate to forward packets certainly), it will check whether it is a neighbor node of
two adjacent listed forwarder pair in the forwarder list. This work can be done easily by a
searching algorithm with complexity O(n × m), where n is the number of neighbors of A or
B, and m is the length of the forwarder list.
In our example shown in Figure 13.8, both A and B are the neighbors of X and Y, and
we should evaluate further like follows. Take node A as an example: it will check whether
R(A, X) is greater than R(Y, X) and whether R(A, Y) is greater than R(X, Y). If we assume that
links are symmetric, R(X, Y) should equal R(Y, X), and four such RSSI values are maintained
in the Neighbor Table. If both of the previous two comparisons are true, node A knows it is
a valid small-scale retransmitter but cannot guarantee it is the best one. To check whether it
270  Opportunistic Networks

is the best small-scale retransmitter, it will look up whether other valid small-scale retrans-
mitters exist in the network. This can be done by running a simple searching algorithm on
the listed forwarders’ RSSI Tables. In our example, we should search the same Neighbor’s
Neighbor ID from X’s and Y’s RSSI Tables in A’s Neighbor Table; and the complexity is
O(nX × nY), where nX and nY are the number of neighbors of nodes X and Y. In our example,
another valid small-scale retransmitter would be B if both R(B, X) is greater than R(Y, X) and
R(B, Y) is greater than R(X, Y) are true.
If B is also a valid small-scale retransmitter judged by node A, then node A has to make
a decision regarding which one (A itself or B) is better. We propose a scoring function given
in the following to evaluate the score of every valid small-scale retransmitter i between M
and N.

 1 1 
W i    4 4 
 R  i, N  R  i, M  

  1 1  
 max  ,   (13.5)
1 
 R  i, N  R  i, M    1 1 
   ln  ln  4 4  .
 4  1 1   R  i, N 
 R  i, M   
min  ,
  R  i, N  R  i, M   
   

Hence, A has to compare the score of itself, given by

 1 1 
W  A   4 4 
 R  A, Y  R  A, X  

  1 1   (13.6)
 max  ,  
 R  A, Y  R  A, X  
1
   ln    ln  4 1
4
1 

   R  A, Y  R  A, X   
4 min 
1
,
1
 
  R  A, Y  R  A, X   
   

to the score of node B by given by

 1 1 
W  B   4 4 
 R  B, Y  R  B, X  

  1 1  
 max  ,   (13.7)
 R  B, Y  R  B, X  
1
   ln    ln  4 1
4
1 
 .
 4  1 1   R  B, Y 
 R  B, X   
min  ,
  R  B, Y  R  B, X   
   

and the less one should be chosen as the result of small-scale retransmitter selection algo-
rithm. The same comparison will also be operated on node B, because node B knows node A
is a valid small-scale retransmitter too.
The algorithm does not need an additional packet to coordinate the final decision on dif-
ferent nodes because the RSSI Tables kept for node X and Y on node A and the RSSI Tables
kept for node X and Y on node B must be the same. That is because if A and B are neighbors
 Opportunistic Routing in Mobile Networks  271

for both X and Y, the RSSI information broadcasted from X and Y will be received simul-
taneous by nodes A and B, and both of them will refresh their corresponding RSSI Tables
together. Therefore, every valid small-scale retransmitter will finally agree on a particular one
as the best small-scale retransmitter out of themselves without gossip.
An important explanation must be clarified here, which is why we propose Equation 13.5
to evaluate the valid small-scale retransmitters. Even though the RSSI calculation method
is not specified in the 802.11 standard, 802.11 device vendors should calculate the RSSI
according to the received signal strength, and the higher the received signal strength, the
greater RSSI value should be returned from hardware [45]. Ideally, the path loss model of
wireless communication follows the rule that the received signal power is inversely propor-
tional to dn, where d is the distance between the transmitter and the receiver and n is usually
an integer within 2, 3, and 4. If we take other parameters as constants, the ideal relationship
between distance from M to N and RSSI can be presented as

R  M ,N   C  d M
n
,N (13.8)

where C is a constant.
When we want to get the reciprocal of RSSI, we have

1 dn
 M,N (13.9)
R  M ,N  C

The rationality of comparing scores calculated by Equation 13.5 is explained as follows.

Considering Equation 13.9, the Equation 13.5 can be simplified to:

  n n
 
  di4, M di4, N  
 max ,
 4C 4C  
 n n
     n n
 
 d 4 d 4
  d d
  ln  i , M  i , N  
4 4
W  i    4 i , M  4 i , N    ln  
 C C    n n
 4
 C
4
C 
  min  di , M , di , N 
4 4
   (13.10)
  4C 4C  
   
   
 n n

 di4, M  di4, N 
   n  ln max  di , M ,di , N   4  ln  d 4  d 4   lnC 
n n

  ,N 
 min  di , M ,di , N    
i , M i
4 4 C    

We can see that a node which leads the minimum value of Equation 13.10 should be the
node which has nearly the same distances to both node M and node N, and it should have
the least summation of two such distances. The parameters in Equation 13.5 are selected by
a great deal of trials in MATLAB. Finally, the figures we created in MATLAB when n equals
2, 3, and 4 are shown in Figures 13.10 through 13.12.
In Figures 13.10 through 13.12, the rectangle with the Width/Height ratio 3 is the π/2
rotated circumscribed quadrilateral of the olivary shape in Figure 13.8, which is surrounded by
 and PYQ
PXQ 
 . The nodes X and Y are located on the position 50  3,0 and 50  3,100 ,
  
and the nodes P and Q are located on the position (0, 50) and 100   3,50 . The contour
line is the curve on which all nodes have the same score given by Equation 13.10. We can see
272  Opportunistic Networks

(a) Contour line from top

(b) Profile

Figure 13.10  Scoring function with n = 2 as the path loss parameter.

(a) Contour line from top

(b) Profile

Figure 13.11  Scoring function with n = 3 as the path loss parameter.

 Opportunistic Routing in Mobile Networks  273

(a) Contour line from top

(b) Profile

Figure 13.12  Scoring function with n = 4 as the path loss parameter.

that, for all possible n from 2 to 4, the contour lines follow the same rule that when a node is
nearer to one of two connected neighbors, the node’s score is greater, so it has less opportunity
to be selected. That is reasonable because the small-scale retransmitter which is too near with
one of the two nodes that are connected by a vulnerable link contributes less than that in the
middle of such two nodes. Furthermore, the center point on every figure always has the mini-
mum value, which indicates to us where the best small-scale retransmitter is supposed to be.
For example, P2 in all the aforementioned figures probably has too little of a contribution
than P1 to be a small-scale retransmitter. One more thing: P2 should have a better position
than P4 because P4 is near the position of P or Q in Figure 13.8 and has a longer distance to
both nodes X and Y than to P2, and from our scoring function, it does have a smaller score
than P4. Hence, the contribution made by sender diversity and receiver diversity from P2 is
more than the contribution given by P4. Furthermore, considering P2 is better than P4 and
worse than P1, and P4 and P1 are on the same center line, we predict there must be a point on
the center line which has the same score with P2. In fact, we do find such a point: P3, which
is on the same contour line of P2.
Based on the analysis before, we can see that the scoring function we proposed matches
our predications very well, so we proved the function is suitable to be used to evaluate valid
small-scale retransmitters.

13.5 PERFORMANCE EVALUATION

In this section, we will study the performance of the proposed solution that enables the
opportunistic data forwarding in MANETs. In particular, we study not only the overall per-
formance where all the solutions work together, but also the particular effectiveness of PSR
274  Opportunistic Networks

(Section 13.3) and the small-scale retransmission (Section 13.5). Hence, this part consists
of the following three subsections: the performance study of PSR, the effectiveness study of
small-scale retransmission, and the overall performance of CORMAN.

13.5.1 Performance Study of PSR

We study the performance of PSR using computer simulation with Network Simulator 2 (ver-
sion 2.34). We compare PSR against OLSR [4], DSDV [3], and DSR [5], three fundamentally
different routing protocols in MANETs, with varying network densities and node mobil-
ity rates. We measure the data transportation capacity of these protocols supporting TCP
(Transmission Control Protocol) and UDP (User Datagram Protocol) with different data flow
deployment characteristics. Our tests show that the overhead of PSR is indeed only a fraction
of that of the baseline protocols. Nevertheless, as it provides global routing information at
such a small cost, PSR offers similar or even better data delivery performance. In this section,
we first describe how the experiment scenarios are configured and what measurements are
collected. Then we present and interpret the data collected from networks with heavy TCP
flows and from those with light UDP streams.

13.5.1.1 Experiment Settings
We use the Two-Ray Ground Reflection channel propagation model in our simulation.
Without loss of generality, we select a 1 Mbps nominal data rate at the 802.11 links to
study the relative performance among the selected protocols. With the default Physical Layer
parameters of the simulator, the transmission range is approximately 250 m and the carrier
sensing range is about 550 m.
We compare the performance of PSR with that of OLSR, DSDV, and DSR. We selected
these baseline protocols that are different in nature for the following reasons. On one hand,
OLSR and DSDV are both proactive routing protocols, and PSR is also in this category. On
the other hand, OLSR makes a complete topological structure available at each node, whereas
in DSDV, nodes have only distance estimates to other nodes via a neighbor. PSR sits in the
middle ground, where each node maintains a spanning tree of the network. Furthermore,
DSR is a well-accepted reactive source-routing scheme and, as with PSR, it supports source
routing, which does not require other nodes to maintain forwarding lookup tables. For a
more leveled comparison to the peer proactive protocols, we make PSR transport regular IP
packets rather than source-routed packets, although it can and should be used as a source-
routing scheme. All three baseline protocols are configured and tested out of the box of NS-2.
In modeling node mobility of the simulated MANETs, we use the Random Waypoint
Model to generate node trajectories. In this model, each node moves towards a series of
target positions. The rate of velocity for each move is uniformly selected from [0, vmax]. Once
it has reached a target position, it may pause for a specific amount of time before moving
towards the next position. All networks have 50 nodes in our tests. We have two series of
scenarios based on the mobility model. The first series of scenarios has a fixed vmax but dif-
ferent network densities by varying the network dimensions. The second series has the same
network density but varying vmax.
We study the data transportation capabilities of these routing schemes and their overhead
in doing so by loading the networks with TCP data flows and UDP voice streams.

• To test how TCP is supported, in each scenario, we randomly select 40 nodes out of the
50 and pair them up. For each pair, we set up a permanent one-way FTP (File Transfer
Protocol) data transfer. We repeat the selection of the 40 nodes five times and study
 Opportunistic Routing in Mobile Networks  275

their collective behavior. This essentially mimics heavily loaded mobile networks. For
all four protocols, we measure their TCP throughput, end-to-end delay, and routing
overhead in byte per node per second in each scenario.
• To study their performance in supporting UDP, we use two-way Constant Bit Rate
(CBR) streams for compressed voice communications. Specifically, we select three pairs
of nodes and feed each node with a CBR flow of 160 byte/pkt and 10 pkt/s, which simu-
lates mobile networks with a light voice communication load. We measure the Packet
Delivery Ratio (PDR), end-to-end delay, and delay jitter in each scenario.Results about
TCP (Sections 13.5.1.2 and 13.5.1.3) and UDP (Sections 13.5.1.4 and 13.5.1.5) with
regard to varying node densities and velocity rates are in the subsequent subsections.

13.5.1.2 TCP with Node Density

We first study the performance of PSR, OLSR, DSDV, and DSR in supporting 20 TCP flows
in networks with different node densities. Specifically, with the default 250 m transmission
range in ns-2, we deploy our 50-node network in a square space of varying side lengths that
yield node densities of approximately 5, 6, 7, …, 12. These nodes move following the random
waypoint model with vmax = 30 m/s.
We plot in Figure 13.13 the per-node per-second routing overhead, i.e., the amount of
routing information transmitted by the routing agents measured in byte/node/second, of the
four protocols when they transport a large number of TCP flows. This figure shows that the
overhead of PSR (20 to 30) is just a fraction of that of OLSR and DSDV (140 to 260), and
more than an order of magnitude smaller than DSR (420 to 830). The routing overhead of
PSR, OLSR, and DSDV goes up gradually as the node density increases. This is a typical
behavior of proactive routing protocols in MANETs. Such protocols usually use a fixed time
interval to schedule route exchanges. While the number of routing messages transmitted in
the network is always constant for a given network, the size of such a message is determined

PSR
OLSR
DSDV
800 DSR
750
700
Overhead (bytes/node/second)

650
600
550
500
450
400
350
300
250
200
150
100
50
0
5 6 7 8 9 10 11 12
Density

Figure 13.13  Routing overhead vs. density.

276  Opportunistic Networks

by the node density. That is, a node periodically transmits a message to summarize changes
as nodes have come into or gone out of its range. As a result, when the node density is
higher, a longer update message is transmitted even if the rate of node motion velocity is the
same. Note that when the node density is really high, say around 10 and 12, the overhead of
OLSR flattens out or even slightly decreases. This is a feature of OLSR when its Multi-Point
Relay mechanism becomes more effective in removing duplicate broadcasts, which is the
most important improvement of OLSR over conventional link-state routing protocols. PSR
uses a highly concise design of messaging, allowing it to have a much smaller overhead than
the baseline protocols. In contrast, DSR as a reactive routing protocol incurs a significantly
higher overhead when transporting a large number of TCP flows because every source node
needs to conduct its own route search. This is not surprising, as reactive routing protocols
were not meant to be used in such scenarios. Later in our experiments (Sections 13.5.1.4
and 13.5.1.5), we test all four protocols supporting just a few UDP streams for a different
perspective. Here, the routing overhead of DSR decreases with the node density going up
and network diameter going down. This is because the number of hops to a destination is
smaller in a denser network, so the shorter, more robust routes break less frequently and do
not need as many route searches. Furthermore, compared to IP forwarding, the fact that DSR
is source routing and that intermediate nodes cannot modify the routes embedded in data
packets works against its performance in a mobile network, both in terms of the increase of
search operations and the loss of data transportation capacity. The reason is that, because
a source node can be quite a few hops away from the destination, its knowledge about the
path as embedded in the packets can quickly become obsolete in a highly mobile network. As
a packet progresses en route, if an intermediate node cannot reach the next hop as indicated
in the embedded path, it will be dropped. This is very different from IP forwarding, where
intermediate nodes can have more updated routing information than the source and can uti-
lize that information in forwarding decisions.
Figure 13.14 plots the TCP throughput of the four protocols for the same node density
levels as before. The total throughput of the 20 TCP flows of PSR, OLSR, and DSDV is notice-
ably higher than that of DSR. In addition, while the TCP throughput of DSR decreases with
node density, that for the other three are somewhat unaffected, hovering at around 500 kbps.
Apparently, the large routing overhead of DSR, especially in dense networks, consumes a fair
amount of channel bandwidth, leaving less room for data transportation. In most cases, PSR
has the highest throughput because it needs to give up the least network resources for routing.
Next, we focus on the end-to-end delay of TCP flows to investigate how well these proto-
cols support time-sensitive applications. Figure 13.15 presents the delay measured for differ-
ent node densities. As the density increases from 5 to 12 neighbors, the delay of DSR goes up
from 0.58 to about 1.5 s, which is significantly higher than the typical value of 0.15 to 0.35
s for the other three protocols. Such a difference is caused by the initial route search when
a TCP flow starts, and by the subsequent searches triggered by route errors. As the network
becomes denser, all protocols show an increasing trend in end-to-end delay. This may seem
counter-intuitive, as in denser networks, the average hop distance between source-destination
pairs is smaller, which should lead to shorter round-trip time. However, this benefit is com-
pletely offset by more intense channel contention. Recall that the node density is inversely
proportional to the square of network diameter. As such, in the interplay between route
length and channel contention, the latter dominates the overall effect.

13.5.1.3 TCP with Velocity

We also study the performance of PSR and compare it to OLSR, DSDV, and DSR with dif-
ferent rates of node velocity. In particular, we conduct another series of tests in networks
 Opportunistic Routing in Mobile Networks  277

PSR
OLSR
650 DSDV
DSR
600
550
500
Throughput (kbps)

450
400
350
300
250
200
150
100
50
0
5 6 7 8 9 10 11 12
Density

Figure 13.14  TCP throughput vs. density.

PSR
OLSR
DSDV
DSR
1.3
1.2
1.1
End-to-end delay (s)

1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
5 6 7 8 9 10 11 12
Density

Figure 13.15  End-to-end delay in TCP vs. density.

of 50 nodes deployed in a 1100 × 1100 (m2) square area with vmax set to 0, 4, 8, 12, … , 32
(m/s). The network thus has an effective node density of around 7 neighbors per node, i.e.,
a medium density among those configured in the previous subsection. As before, 20 TCP
one-way flows are deployed between 40 nodes, and we measure the routing overhead, TCP
throughput, and end-to-end delay (Figures 13.16 through 13.18).
The routing overhead of all four protocols with varying rates of node velocity is plotted as
in Figure 13.16. Note that the velocity to the right of the x-axis corresponds to the middle bars
in Figure 13.13. We observe in the plot here, as vmax decreases, the overhead of all protocols
278  Opportunistic Networks

PSR
OLSR
DSDV
DSR
750
700
Overhead (bytes/node/second)

650
600
550
500
450
400
350
300
250
200
150
100
50
0
0 4 8 12 16 20 24 28 32
vmax (m/s)

Figure 13.16  Routing overhead vs. velocity.

comes down. The reason for DSR is that, as the network structure becomes more stable, fewer
route repair attempts are necessary. For the case of the proactive protocols, it is the reduction
in the size of routing messages (i.e., fewer neighbors have changed positions) that cuts down
the overhead. Still relative among these four protocols, when the network is not stationary
(vmax ≠ 0), the overhead of PSR (20 to 30 byte/node/second) is a fraction of that of OLSR and
DSDV (90 to 300), and more than an order of magnitude lower than DSR (180 to 770).
The TCP throughput and end-to-end delay are plotted in Figures 13.17 and 13.18 respec-
tively. From these figures, we observe that the performance of PSR, OLSR, and DSDV are

PSR
OLSR
DSDV
900 DSR

800

700
Throughput (kbps)

600

500

400

300

200

100

0
0 4 8 12 16 20 24 28 32
vmax (m/s)

Figure 13.17  TCP throughput vs. velocity.

 Opportunistic Routing in Mobile Networks  279

PSR
OLSR
DSDV
DSR
1.1

0.9
End-to-end delay (s)

0.8

0.7
0.6

0.5

0.4

0.3

0.2

0.1

0
0 4 8 12 16 20 24 28 32
vmax (m/s)

Figure 13.18  End-to-end delay in TCP vs. velocity.

similar, with PSR leading the pack in most cases. In addition, neither throughput or delay is
affected by the different rates of velocity. The only exception is that when vmax = 0, all proto-
cols yield a high throughput of 900 kbps. With a greater portion of the channel bandwidth
devoured by routing messages in highly mobile networks, DSR suffers a noticeable perfor-
mance penalty in TCP throughput and end-to-end delay.

13.5.1.4 UDP with Density

We also tested the four protocols for their performance in transporting a small number of
UDP streams. This is a typical assumption for ideal scenarios of reactive routing protocols.
Here, we deploy three two-way UDP streams in order to simulate compressed voice communi-
cations. To find out about how node density affects these protocols, we use the same network
and mobility configurations as in Section 13.5.1.2. We measure and plot the PDR (Figure
13.19), delay (Figure 13.20), and delay jitter (Figure 13.21) against varying node densities.
In Figure 13.19, the PDRs of all four protocols are in the same ballpark across different
node densities, with DSR slightly in the lead and OLSR trailing behind. This verifies that the
traffic configuration is favorable for DSR. The relatively high loss rate of OLSR among the
proactive routing protocols is caused by the higher routing overhead compared to PSR and
DSDV. When the nodes are neither too sparse, so that the network connectivity is good, nor
too dense, so that the channel can be spatially reused, these protocols have a fairly high PDR
of over 70% for PSR, DSDV, and DSR, and 60% to 70% for OLSR.
When we turn to end-to-end delay (Figure 13.20), there is a noticeable difference between
DSR and the proactive protocols. In particular, DSR as reactive protocol has a rather large
delay in sparse networks. This is because the long, vulnerable routes discovered during the
search procedure break frequently, forcing nodes to hold packets back for an extended period
before new routes are identified. Oppositely, the network sparsity does not affect proactive
protocols as much because their periodic routing information exchange makes them more
prepared for network structure alteration. While the delay of DSR is off the chart, that of PSR
is always less than 0.05 s, which is also much less than that for DSDV and OLSR (0.1–0.43 s).
280  Opportunistic Networks

PSR
OLSR
DSDV
DSR
90

80
Packet delivery ratio (%)

70

60

50

40

30

20

10

0
5 6 7 8 9 10 11 12
Density

Figure 13.19  PDR in UDP vs. density.

PSR
OLSR
3.5 DSDV
DSR

3
End-to-end delay (s)

2.5

1.5

0.5

0
5 6 7 8 9 10 11 12
Density

Figure 13.20  End-to-end delay in UDP vs. density.

On a related note, the delay jitter (Figure 13.21) of PSR is significantly lower than the other
three. Note that Voice-Over-IP (VoIP) applications usually discard packets that arrive too
late. Therefore, the jitter among the packets actually used by the VoIP receiving agent is much
smaller. Nevertheless, our metric still reflects how consistent these protocols are in delivering
best-effort packets.
 Opportunistic Routing in Mobile Networks  281

PSR
OLSR
16 DSDV
15 DSR
14
13
End-to-end delay variance (s)

12
11
10
9
8
7
6
5
4
3
2
1
0
5 6 7 8 9 10 11 12
Density

Figure 13.21  End-to-end delay jitter in UDP vs. density.

13.5.1.5 UDP with Velocity

The same measurements are taken to test these protocols in response to different rates of
node velocity. As with the case of the previous subsection, we pick three node pairs out of the
50 nodes and give them two-way CBR streams.
For the entire series of different velocity caps vmax = 0, 4, 8, 12,…, 32 m/s, the node density
is again set to around 7 neighbors per node.
From the plot of PDR (Figure 13.22), we observe that DSR is able to support three voice
streams with little packet loss. Specifically, the PDR of DSR, PSR, and DSDV is always over
70% even when vmax = 32. The reliability of OLSR is relatively lower, which can go below
60% at high speed (vmax = 28 or 32 m/s). Note that all four protocols are very reliable in data
delivery when vmax = 0 or 4 m/s, where the loss rates are well below 10%. Their performance
in terms of PDR degrades gracefully as the rate of node velocity increases.
The end-to-end delay (Figure 13.23) presents a rather distinct landscape. In particular,
the number for DSR is significantly higher than the other protocols except in low-mobility
networks with vmax = 0 or 4 m/s. In all cases, the delay for PSR is much smaller compared to
OLSR and DSDV. On the other hand, the measured delay jitter (Figure 13.24) indicates that
all protocols become less consistent when nodes move faster. Relatively speaking, however,
the variance of PSR is much smaller than the other three.

13.5.2 Effectiveness Study of Small-Scale Retransmission

In this section, we study the effectiveness of small-scale retransmission in particular by run-
ning computer simulation using NS-2. To investigate the effectiveness of small-scale retrans-
mission, we test CORMAN with small-scale retransmission enabled and CORMAN with
small-scale retransmission disabled separately. The performance improvement and explana-
tions of the simulation results are explained in the rest of this section.
282  Opportunistic Networks

PSR
OLSR
DSDV
DSR
100

90
Packet delivery ratio (%)

80

70

60

50

40

30

20

10

0
0 4 8 12 16 20 24 28 32
vmax (m/s)

Figure 13.22  PDR in UDP vs. velocity.

PSR
OLSR
DSDV
DSR
1.4

1.2
End-to-end delay (s)

1.0

0.8

0.6

0.4

0.2

0
0 4 8 12 16 20 24 28 32
vmax (m/s)

Figure 13.23  End-to-end delay in UDP vs. velocity.

13.5.2.1 Experiment Settings
The channel propagation model used in NS-2 had been predominantly the Two-Ray Ground
Reflection model early on. However, this model is realized to be a simplified path loss model
without considering fading. In our work, we choose the Nakagami propagation model to
test CORMAN in a more realistic fading environment. The probability density function of
Nakagami distribution of the received signal’s amplitude X = r(r ≥ 0) is defined as:

2    
f  r,  ,    r 2  1 exp   r 2  , (13.11)
       
 Opportunistic Routing in Mobile Networks  283

PSR
OLSR
DSDV
DSR
10

9
End-to-end delay variance (s)

0
0 4 8 12 16 20 24 28 32
vmax (m/s)

Figure 13.24  End-to-end delay jitter in UDP vs. velocity.

E 2  X 2 
where   and ω = E[X2]. In NS-2, when a node has received a packet, it first cal-
Var  X 2 
culates the received power using path loss based on the Frii Free-Space model. This value is
compensated with Nakagami’s fluctuation before further processing. We configure the nomi-
nal data rate at the 802.11 links to 1 Mbps. In modeling node motion, we also adopt the
Random Waypoint model (Section 13.5.1.1) to generate the simulation scenarios.
We inject CBR (constant bit rate) data flows in the network, which are carried by UDP.
Specifically, a source node generates 50 packets every second, each has a payload of 1000
bytes. This translates to a traffic rate of 400 kbps injected by a node. When comparing
different CORMAN versions which has and does not have the small-scale retransmission
functionality, we record the packet delivery ratio (PDR), i.e. the fraction of packets received
by the destination out of all the packets injected, and end-to-end delay average and variance.
We observe that the CORMAN with small-scale retransmission enabled outperforms the
CORMAN with small-scale retransmission disabled in terms of all of these metrics.

13.5.2.2 Performance versus Network Dimension

We make performance comparison of CORMAN under different configurations. One is with
the small-scale retransmission enabled and the other is with such a module disabled, and our
tests are against the varying of the network density. In particular, we have network tomog-
raphies of l × l (m2), where l = 450, 500, 550, …, 950. We deploy 50 nodes in each of these
network dimensions to test the protocols with differing node densities, and every node moves
randomly with the Random Waypoint model at vmax = 20 m/s. For each dimension scenario,
we test performances of CORMAN with small-scale retransmission enabled and CORMAN
with small-scale retransmission disabled in transporting CBR data flows between a randomly
selected source-destination pair. We repeat this process 20 times for a given scenario. We mea-
sure the PDR, end-to-end delay, and delay jitter for both protocols and average them over
the 20 repetitions of each scenario, as plotted in Figures 13.25 through 13.27, respectively.
284  Opportunistic Networks

Disabled
Enable
80

70
Packet delivery ratio (%)

60

50

40

30

20

10

0
450 500 550 600 650 700 750 800 850 900 950
Network dimension (m)

Figure 13.25  PDR vs. network dimension.

Disabled
Enable
2.50

2.25

2.00
End-to-end delay (s)

1.75

1.50

1.25

1.00

0.75

0.50

0.25

0.00
450 500 550 600 650 700 750 800 850 900 950
Network dimension (m)

Figure 13.26  Packet delay vs. network dimension.

We observe that when the network density is relatively low, CORMAN with small-scale
retransmission disabled has better PDR. In particular, when the side length of the square net-
work boundary grows from 450 to 500 m, the CORMAN with disabled small-scale retrans-
mission outperform the CORMAN with the proposed scheme enabled. When the side length
of the square boundary kept increasing from 500 m, the CORMAN with small-scale retrans-
mission enabled outperform its opposite, and the greater the side length, the more obvious
PDR gain could be achieved.
 Opportunistic Routing in Mobile Networks  285

Disabled
Enable

End-to-end delay variance (s) 3.5

2.5

1.5

0.5

0
450 500 550 600 650 700 750 800 850 900 950
Network dimension (m)

Figure 13.27  Delay jitter vs. network dimension.

To investigate the reason behind such a phenomenon, we studied the simulation trace
files deeply and plotted out the number of packets forwarded by local relay nodes and the
collisions on listed forwarders caused by local relay nodes in Figure 13.28. The left scale of
Figure 13.28 gives us the number of packets forwarded by small-scale retransmitters and
the collisions on listed forwarders caused by small-scale retransmitters, and the right scale
gives us the ratio of the two values. It is obvious that both the small-scale forwarding and
the collisions they caused increase when the network dimension goes up. Furthermore, the

Forwardings count
Collisions count
Forwardings over collisions
Related events count (forwardings and collisions)

3000

2750
1.0
2500
Forwardings over collisions

2250 0.9
2000
1750 0.8
1500
0.7
1250

1000
0.6
750

500 0.5
250

0
450 500 550 600 650 700 750 800 850 900 950
Network dimension (m)

Figure 13.28  Reasons analysis of PDR reversion.

286  Opportunistic Networks

forwarding count increases faster than the collisions they caused. In our further investiga-
tion, we found the forwarding increases because the hops from source to destination node
increase with the network dimensions. However, we found the increases of collisions are not
due to the growth of hop count because we have granted nodes the priority to transmit the
packets in the same batch [9]. The real reason is that a route with more hops from source to
destination reduces the size of fragment transmitted by intermediate node, so the entire frag-
ment has a higher probability of being lost on both listed forwarders and local relay nodes.
As the result, nodes cannot predict the exact time to start their transmission, and that is the
real reason for more collisions in the network. Therefore, the collision will not grow directly
against the increases of network dimensions and with a lower growth speed than that the
increasing of forwarding times.
In a summary, because wireless links in a dense network have relatively high reliability
with relatively less hop count – note that the nodes have to try to deliver a same packet sev-
eral times before discarding them – so the contribution of small-scale retransmission in dense
network is in fact limited. When the density of the network decreases, the benefit achieved by
using small-scale retransmission shows its advantage.
From Figure 13.26 we can see that the packet end-to-end delay in both tested schemes with
and without small-scale retransmission are nearly the same in a dense network. The scheme
with local cooperative relay enabled has shorter packet end-to-end delay in a sparse network,
because the local cooperative relay will save packets and prevent them from being delivered
from original source again. The Figure 13.27 indicate that both the CORMANs with and
without small-scale retransmission have nearly the same packet end-to-end delay jitter.

13.5.2.3 Performance versus Velocity

We also study CORMAN’s performance with the small-scale retransmission enabled or not
by varying the node velocity. We conduct another set of tests in a network of 50 nodes
deployed in a 700 × 700 (m2) space with a varying vmax, where vmax = 0, 3, 6, …, 30 (m/s).
For each velocity scenario, we test CORMAN with small-scale retransmission enabled and
disabled in transporting CBR data flows between a randomly selected source-destination
pair as well. We repeat this process 20 times for a given scenario. We collect the PDR, end-to-
end delay, and delay jitter for both protocols averaged over each scenario, and plot them in
Figures 13.29 through 13.31, respectively.
From Figure 13.29, we can see that for all velocity scenarios, the small-scale retransmission
enabled scheme outperformed the small-scale retransmission disabled scheme for the same
reasons as we analyzed in Section 13.5.2.2, so we will not explain that again here. What
makes these results interesting is that, even though both the PDRs in two test conditions
decrease together when nodes move faster, the PDR gain by using small-scale retransmission
in CORMAN grows when the velocity goes up. That is due to the scoring function in the
small-scale retransmitter being a real-time and distributed algorithm, and it can provide the
best retransmitter on time to reduce the effect of nodes’ mobility. As a result, the small-scale
retransmission has more effectiveness to fix up the break links when the node velocities go up.
Also, CORMAN with small-scale retransmission enabled has shorter end-to-end delay
and almost same end-to-end delay jitter compared to that with small-scale retransmission
disabled, due to the same reasons we analyzed in Section 13.5.2.2.

13.5.3 Overall Performance of CORMAN

In this section, we study the performance of CORMAN by running computer simulation
using NS-2 (version 2.34). We compare it against AODV with varying network densities and
 Opportunistic Routing in Mobile Networks  287

Disabled
Enable
80

70
Packet delivery ratio (%)

60

50

40

30

20

10

0
0 3 6 9 12 15 18 21 24 27 30
vmax (m/s)

Figure 13.29  PDR vs. node velocity.

Disabled
Enable
2.50

2.25

2.00
End-to-end delay (s)

1.75

1.50

1.25

1.00

0.75

0.50

0.25

0.00
0 3 6 9 12 15 18 21 24 27 30
vmax (m/s)

Figure 13.30  Packet delay vs. node velocity.

node mobility rates. The performance improvement and explanations of these results are
explained in the rest of this section.

13.5.3.1 Experiment Settings
The basic configurations of the simulations for the overall performance in CORMAN are
nearly the same as that in the simulations for the effectiveness of small-scale retransmission
in Section 13.5.2, where we choose the Nakagami propagation model to test CORMAN’s
overall performance, configure the nominal data rate at the 802.11 links to 1 Mbps, and
288  Opportunistic Networks

Disabled
Enable

End-to-end delay variance (s) 3.5

3.0

2.5

2.0

1.5

1.0

0.5

0
0 3 6 9 12 15 18 21 24 27 30
vmax (m/s)

Figure 13.31  Delay jitter vs. node velocity.

adopt the Random Waypoint mobility model to generate the simulation scenarios. As well,
we inject CBR data flows in the network carried by UDP with the same data rate as we
specified in Section 13.5.2. We compare the overall performance of CORMAN with that
of AODV [6]. We select AODV as the baseline because AODV is a widely adopted routing
protocol in MANETs, and its behavior in both NS-2 and real network operations is well
understood by the research community.

13.5.3.2 Performance versus Network Dimension

We first compare the performance of CORMAN and AODV with different network dimen-
sions. Specifically, we have network tomographies of l × l(m2), where l = 250, 300, 350, …,
1000. We deploy 50 nodes in each of these network dimensions to test the protocols with
differing node densities. These nodes move following the Random Waypoint model with
vmax = 10 m/s. For each dimension scenario, we test CORMAN and AODV’s capabilities in
transporting CBR data flows between a randomly selected source-destination pair. We repeat
this process five times for a given scenario. We measure the PDR, end-to-end delay, and delay
jitter for both protocols and average them over the five repetitions of each scenario, as plot-
ted in Figures 13.32 through 13.34, respectively.
We observe that CORMAN has a PDR (Figure 13.32) of about 95% for dense networks
(i.e., 250 ≤ l ≤ 500 m). As the node density decreases, this rate gradually goes down to
about 60%. In contrast, AODV’s PDR ranges between 60% and 80% for dense networks
and quickly drops to around 20% for sparse networks. (We use a red plotting series to indi-
cate the relative performance of CORMAN over PDR in all of our figures.) There are two
reasons for the PDR penalty for AODV to operate in sparse networks. First, data packets
are forwarded using traditional IP forwarding in AODV. When channel quality varies (as
emulated by the Nakagami model), a packet may be lost at the link layer. After a few failed
retransmits, it will be dropped by the network layer. CORMAN, however, is designed to
utilize such link effects so that at least one downstream node would be available despite the
link variation. CORMAN facilitates opportunistic data forwarding using the link quality
 Opportunistic Routing in Mobile Networks  289

CORMAN
AODV
Relative
100

Relative performance (CORMAN vs. AODV %)

400
90

80 350

70 300
PDR (%)

60 250
50
200
40
150
30
100
20

10 50

0 0
25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

10
0

00
Network dimension (m)

Figure 13.32  PDR vs. network dimension.

CORMAN
AODV
Relative
0.9

Relative performance (CORMAN vs. AODV %)

180
0.8
160
0.7
140
Average delay (s)

0.6
120
0.5
100

0.4 80

0.3 60

0.2 40

0.1 20

0 0
25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

10
0

00

Network dimension (m)

Figure 13.33  Packet delay vs. network dimension.

diversity at different receivers and allows them to cooperate with each other with a minimal
overhead. Consequently, CORMAN has a strong resilience to link quality fluctuation and
node mobility. Second, the route search of AODV does not function well with unreliable
links. Recall that, in AODV, when a node finds that it does not have a next hop available for
a given data packet, it broadcasts a RREQ (route request) to find one. Both the destination
and any intermediate node that has a valid cached route can reply with a RREP (route reply).
When links were perfectly symmetric, the RREP packet would take the inverse path leading
290  Opportunistic Networks

CORMAN
AODV
Relative

Relative performance (CORMAN vs. AODV %)

1.1 200
1.0
180
0.9
160
Delay variance (s)

0.8
140
0.7
120
0.6
100
0.5
80
0.4
60
0.3

0.2 40

0.1 20

0 0
25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

10
0

00
Network dimension (m)

Figure 13.34  Delay jitter vs. network dimension.

to the initiating node of the route search. However, when links suffer from fading, the RREP
packets may not be able to propagate back to the initiator because of transient low link qual-
ity in the reverse direction. As a result, it takes AODV much longer to obtain stable routes. In
fact, this performance loss has been observed in earlier studies of AODV when large numbers
of unidirectional links are present in the network [46]. In our NS-2 simulation using the
Nakagami propagation model, the independent link quality fluctuation in both directions
essentially produces temporary unidirectional links. This negative effect of fading links on
AODV can be mediated to a degree when the node density increases. Therefore, in our next
set of tests (Section 13.5.3.3), all simulation scenarios have a fairly higher node density for
AODV to function reasonably well.
We are also interested in the end-to-end delay and its variance of CORMAN and AODV.
Figure 13.33 presents the end-to-end delay of these protocols in different dimension sce-
narios. We see that, when the node density is higher (i.e., 250 ≤ l ≤ 500 m), CORMAN has a
shorter delay than AODV. For sparser scenarios (i.e., 550 ≤ l ≤ 1000 m), CORMAN’s delay
is slightly longer than AODV but comparable. In CORMAN’s implementation, a node deter-
mines that it can no longer contribute to a batch’s progression if it has not seen an update of
the map after 10 retransmits of its fragment. This is similar to 10 retries at the link layer in
IP-style forwarding. (The default retry limit in 802.11 is 7.) Thus, the not-so-short end-to-end
delay of CORMAN is caused by the larger number of data retransmits. This is a relatively
small cost to pay for a significantly higher PDR (Figure 13.32). The delay jitter (standard
deviation) measured for CORMAN and AODV has similar relative performance for these
scenarios as in Figure 13.34. This is also because of the larger retry limit of CORMAN (10
times) compared to AODV over 802.11 (7 times).

13.5.3.3 Performance versus Velocity

We also study CORMAN’s performance and compare it to AODV in different rates of node
velocity. We conduct another set of tests in a network of 100 nodes deployed in a 300 ×
300 (m2) space with a varying vmax, where vmax = 0, 2, 4,  6, …, 20 (m/s). For each velocity
 Opportunistic Routing in Mobile Networks  291

scenario, we test CORMAN and AODV’s performance in transporting CBR data flows again
between a randomly selected source-destination pair. We repeat this process five times for a
given scenario. We collect the PDR, end-to-end delay, and delay jitter for both protocols aver-
aged over each scenario, and plot them in Figures 13.35 through 13.37, respectively. Note
that these are fairly dense networks, so that AODV has a reasonably high PDR. Since the
network diameter is rather small in this case, the measurements are fairly consistent across
these different velocity scenarios.

CORMAN
AODV
Relative

Relative performance (CORMAN vs. AODV %)

100 200

90 180

80 160

70 140
PDR (%)

60 120

50 100

40 80

30 60

20 40

10 20

0 0
0 2 4 6 8 10 12 14 16 18 20
vmax (m/s)

Figure 13.35  PDR vs. node velocity.

vmax (m/s)

Figure 13.36  Packet delay vs. node velocity.

292  Opportunistic Networks

CORMAN
AODV
Relative

Relative performance (CORMAN vs. AODV %)

0.4 50

0.35 45

40
0.3
Delay variance (s)

35
0.25
30

0.2 25

0.15 20

15
0.1
10
0.05
5

0 0
0 2 4 6 8 10 12 14 16 18 20
vmax (m/s)

Figure 13.37  Delay jitter vs. node velocity.

From Figure 13.35, we observe that CORMAN’s PDR is constantly around 95%, while
that of AODV varies between 57% and 82%. With this very high network density, AODV’s
route search succeeds in a majority of the cases. Yet, it still does not have the same level
of robustness against link quality changes as CORMAN. Compared to AODV, CORMAN
has only a fraction of the end-to-end delay and variance (Figures 13.36 and 13.37) for two
reasons. First, the opportunistic data forwarding scheme in CORMAN allows some packets
to reach the destination in fewer hops than AODV. Second, the proactive routing (PSR) in
CORMAN maintains full-on route information, whereas AODV still has to search for them
if a route is broken. Although route search in AODV usually succeeds in dense networks with
fluctuating link quality, the delay introduced by this process is inevitable.
Based on these observations, CORMAN can maintain its performance despite a high
rate of node velocity with realistic channels emulated by the Nakagami model. Hence, it is
very suitable for many mobile ad hoc network applications, e.g., vehicular ad hoc network
(VANET) applications.

13.6 CONCLUDING REMARKS

13.6.1 Conclusions
In this chapter, we have proposed CORMAN as an opportunistic routing scheme for mobile
ad hoc networks. CORMAN is composed of three components. (1) PSR – a proactive source
routing protocol, (2) large-scale live update of forwarder list, and (3) small-scale retrans-
mission of missing packets. All of these explicitly utilize the broadcasting nature of wire-
less channels and are achieved via efficient cooperation among participating nodes in the
network. Essentially, when packets of the same flow are forwarded, they can take different
paths to the destination. For example, in Figure 13.38, the route between nodes X and Y as
determined by the routing module is indicated by the yellow band. The solid circles repre-
sent the listed forwarders, and the hollow ones are the small-scale retransmitters. In actual
operations of CORMAN, packets p1, p2, and p3 can take separate routes around this band
 Opportunistic Routing in Mobile Networks  293

p2
p3

X p1 Y

Figure 13.38  Packet trajectories.

depending on the transient link quality in the network. Such a decision is made on a per-hop
and per-packet basis. Through computer simulation, CORMAN is shown to have superior
performance measured in PDR, delay, and delay jitter.

13.6.2 Discussion
13.6.2.1 Proactive Source Routing Related
In particular, the PSR is motivated by the need of supporting opportunistic data forward-
ing in mobile ad hoc networks. In order to generalize the milestone work of ExOR for it to
function in such networks, we needed a proactive source routing protocol. Such a protocol
should provide more topology information than just distance vectors but has a significantly
smaller overhead than link-state routing protocols; even the MPR technique in OLSR would
not suffice this need. Thus, we put forward a tree-based routing protocol, PSR, inspired by
PFA and WRP. Its routing overhead per time unit per node is in the order of the number of
the nodes in the network as with DSDV, but each node has the full-path information to reach
all other nodes. For it to have a very small footprint, PSR’s route messaging is designed to be
very concise. First, it uses only one type of message, i.e., the periodic route update, both to
exchange routing information and as hello beacon messages. Second, rather than packaging
a set of discrete tree edges in the routing messages, we package a converted binary tree to
reduce the size of the payload by about half. Third, we interleave full dump messages with
differential updates so that, in relatively stable networks, the differential updates are much
shorter than the full dump messages. To further reduce the size of the differential updates,
when a node maintains its routing tree as the network changes, it tries to minimize the altera-
tion of the tree. As a result, the routing overhead of PSR is only a fraction or less compared
to DSDV, OLSR, and DSR as evidenced by our experiments. Yet, it still has similar or better
performance in transporting TCP and UDP data flows in mobile networks of different veloc-
ity rates and densities.
In the simulation in this work, we used PSR to support traditional IP forwarding for a
closer comparison with DSDV and OLSR, while DSR still carried source-routed messages. In
our simultaneous work, CORMAN [38], we tested PSR’s capability in transporting source-
routed packets for opportunistic data forwarding, where we also found that PSR’s small
overhead met our initial goal. That being said, as indicated in Section 13.5.1.2 earlier, while
alleviating forwarding nodes from table lookup, DSR’s source routing is especially vulner-
able in rapidly changing networks. The reason is that, as a source-routed packet progresses
further from its source, the path carried by the packet can become obsolete, forcing an inter-
mediate node that cannot find the next hop of the path to drop the packet. This is funda-
mentally different from traditional IP forwarding in proactive routing with more built-in
adaptivity, where the routing information maintained at nodes closer to the destination is
often more updated than the source node. Although out of the scope of this research, it
would be interesting to allow intermediate nodes running DSR to modify the path carried
by a source-routed packet for it to use its more updated knowledge to route data to the
294  Opportunistic Networks

destination. This is in fact exactly what PSR does when we used it to carry source-routed
data in CORMAN. Granted, this opens up an array of security issues, which themselves are
part of a vast research area.
As with many protocol designs, in many situations working on PSR, we faced tradeoffs
of sorts. Striking such balances not only gave us the opportunity to think about our design
twice, but also made us understand the problem at hand better. One particular example is
related to trading computational power for data transfer performance. During one route
exchange interval, a node receives a number of routing messages from its neighbors. It needs
to incorporate the updated information to its knowledge base and share it with its neighbors.
The question is when should these two events happen. Although incorporating multiple trees
at one time is computationally more efficient, we chose to do that immediately after receiv-
ing an update from a neighbor. As such, the more accurate information takes effect without
any delay. Otherwise, when a data packet is forwarded to a neighbor that no longer exists, it
causes link layer retrial, backlogging of subsequent packets, and TCP congestion avoidance
and retransmission. With the broadcast and shared nature of the wireless channel, the afore-
mentioned effects are contrary to all other data flows in the area. Therefore, in research on
multi-hop wireless networking, it almost always makes sense for us to minimize any impact
on the network’s communication resources even if there is penalty in other aspects. When
it comes to when a node should share its updated route information with its neighbors,
we chose to delay it until the end of the cycle so that only one update is broadcast in each
period. If a node were to transmit it immediately when there is any change to its routing
tree, it would trigger an explosive chain reaction, and the route updates would overwhelm
the network. As we found out in our preliminary tests, this is the primary reason that WRP’s
overhead was significantly higher than the other protocols under study. PSR has just opened
the box for us, and there will be many more things that we would like to investigate about it.

13.6.2.2 About Large-Scale Live Update and Small-Scale Retransmission

The large-scale live update is another way to utilize the broadcast nature in the wireless com-
munication network, especially for the opportunistic data forwarding with proactive routing
protocols in MANETs. In particular, to avoid too much routing overhead that is injected into
the network, the proactive routing protocols can update the topology changes not in an event-
driven manner but in a timer-driven one, which means the routing update should propagate
from destination to source in a periodical way as we did in PSR (Section 13.3). As a result,
the further from the source node to the destination node, the less accurate routing infor-
mation maintained on the source node. Coherently, the Route-Prioritized Contention-based
opportunistic data forwarding (Section 13.2.1.1.2) always wants the intermediate node that
is nearest to the destination to forward the packets first, where much fresher routing infor-
mation is maintained than upstream nodes which will forward the remaining packets later.
Hence, large-scale live update uses the broadcast nature further in the wireless opportunistic
forwarding network, and when a node takes itself as a frontier (Section 13.4.1) of the batch,
it will update the forwarder list and pack the new one in the data packet. Therefore, when
the data packets are delivered towards the destination, the fresher routing information can be
propagated towards the source node more quickly with no additional overhead.
Furthermore, in the small-scale retransmission we proposed in Section 13.5, we broaden
the opportunistic data forwarding further. In wireless communication networks, the tradi-
tional IP forwarding, opportunistic data forwarding, and opportunistic data forwarding with
small-scale retransmission compose an evolution process. In particular, the most fundamental
approach is the IP forwarding, which is initially proposed for the Internet and only one next
hop address marks the intended receiver in the forwarding process. Then, ExOR [9] makes a
 Opportunistic Routing in Mobile Networks  295

group of receiver to be intended by including a forwarder list in the data packet, and it makes
all receivers forward the packet in a prioritized order. Now, our small-scale retransmission
not only grants the forwarding ability to the receivers included in the forwarder list, but
also lets the nodes which are not shown in the forwarder list participate in data forwarding
between two of them. The broadened opportunistic data forwarding takes the broadcast
nature one step further.

13.6.3 Future Work
Research based on CORMAN can be extended in the following interesting ways.

1. It would be informative to further test CORMAN. For example, we can compare

CORMAN to ExOR and IP forwarding in static multi-hop networks with varying link
quality to study their relative capabilities in data transfer. We will also test these data
transfer techniques when multiple simultaneous flows are present, to study how well
they share the network resources. Through these tests, we will be able to further opti-
mize some parameters of CORMAN, such as the retry limit.
2. The coordination among multiple qualified small-scale retransmitters can be achieved
with better measures than RSSI. In particular, if the “suitability” score is based on
transient link quality rather than historic information, we will be able to better utilize
the receiver diversity. Apparently, this would require non-trivial coordination among
these retransmitters, which could challenging especially when we aim for zero extra
overhead.
3. Nodes running CORMAN forward data packets in fragments. When the source and
destination nodes are separated by many hops, it should allow nodes at different seg-
ments of the route to operate simultaneously. That is, a pipeline of data transportation
could be achieved by better spatial channel reuse. The design of CORMAN can be
further improved to address this explicitly. This may involve making the timing node
back off more precisely and tightly, or even devising a completely different coordina-
tion scheme.

The potential of cooperative communication in multi-hop wireless networks is yet to be

unleashed at higher layers, and CORMAN is only an example. PSR has just opened the box
for us, and there will be many more aspects of it that we would like to investigate.

REFERENCES

[1] Imrich Chlamtac, Marco Conti, and J.-N. Liu. Mobile Ad Hoc Networking: Imperatives and
Challenges. Ad Hoc Networks, 1(1):13–64, 2003.
[2] Rajmohan Rajaraman. Topology Control and Routing in Ad Hoc Networks: A Survey. SIGACT
News, 33:60–73, 2002.
[3] Charles E. Perkins and Pravin Bhagwat. Highly Dynamic Destination-Sequenced Distance-
Vector Routing (DSDV) for Mobile Computers. Computer Communication Review, pp. 234–
244, 1994.
[4] Thomas Clausen and Philippe Jacquet. Optimized Link State Routing Protocol (OLSR). RFC
3626, October 2003.
[5] David B. Johnson, Yih-Chun Hu, and David A. Maltz. On the Dynamic Source Routing Protocol
(DSR) for Mobile Ad Hoc Networks for IPv4. RFC 4728, February 2007.
[6] Charles E. Perkins and Elizabeth M. Royer. Ad hoc On-Demand Distance Vector (AODV)
Routing. RFC 3561, July 2003.
296  Opportunistic Networks

[7] Thomas M. Cover and Abbas A. El Gamal. Capacity Theorems for the Relay Channel. IEEE
Transactions on Information Theory, 25(5):572–584, 1979.
[8] Aria Nosratinia, Todd E. Hunter, and Ahmadreza Hedayat. Cooperative Communication in
Wireless Networks. IEEE Communications Magazine, 42(10):74–80, 2004.
[9] Sanjit Biswas and Robert Morris. ExOR: Opportunistic Multi-Hop Routing for Wireless
Networks. In Proceedings of ACM Conference of the Special Interest Group on Data
Communication (SIGCOMM), pp. 133–144, Philadelphia, PA, August 2005.
[10] Peter Larsson. Selection Diversity Forwarding in a Multihop Packet Radio Network with Fading
Channel and Capture. ACM Mobile Computing and Communications Review, 5(4):47–54,
2001.
[11] Eric Rozner, Jayesh Seshadri, Yogita Mehta, and Lili Qiu. Simple Opportunistic Routing Protocol
for Wireless Mesh Networks. In Proceedings of the 2nd IEEE Workshop on Wireless Mesh
Networks (WiMesh), pp. 48–54, September 2006.
[12] Mathias Kurth, Anatolij Zubow, and Jens-Peter Redlich. Cooperative Opportunistic Routing
Using Transmit Diversity in Wireless Mesh Networks. In Proceedings of the 27th IEEE
International Conference on Computer Communication (INFOCOM), pp. 1310–1318, 2008.
[13] Mohammad Naghshvar and Tara Javidi. Opportunistic Routing with Congestion Diversity in
Wireless Multi-Hop Networks. In Proceedings of the 29th IEEE International Conference on
Computer Communication (INFOCOM), pp. 496–500, Piscataway, NJ, 2010. IEEE Press.
[14] Kai Zeng, Zhenyu Yang, and Wenjing Lou. Opportunistic Routing in Multi-Radio Multi-Channel
Multi-Hop Wireless Networks. In Proceedings of the 29th IEEE International Conference on
Computer Communication (INFOCOM), pp. 476–480, Piscataway, NJ, 2010. IEEE Press.
[15] Shengbo Yang, Feng Zhong, Chai Kiat Yeo, Bu Sung Lee, and Jeff Boleng. Position Based
Opportunistic Routing for Robust Data Delivery in MANETs. In Proceedings of the 2009
IEEE Conference on Global Telecommunications (GLOBECOM), pp. 1325–1330, Honolulu,
Hawaii, December 2009.
[16] Zifei Zhong and Srihari Nelakuditit. On the Efficacy of Opportunistic Routing. In Proceedings of
the 4th IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications
and Networks (SECON), pp. 441–450, June 2007.
[17] Szymon Chachulski, Michael Jennings, Sachin Katti, and Dina Katabi. Trading Structure for
Randomness in Wireless Opportunistic Routing. In Proceedings of ACM Conference of the
Special Interest Group on Data Communication (SIGCOMM), pp. 169–180, Kyoto, Japan,
August 2007.
[18] Douglas S. J. De Couto, Daniel Aguayo, John Bicket, and Robert Morris. A High-Throughput
Path Metric for Multi-Hop Wireless Routing. In Proceedings of the 9th Annual International
Conference on Mobile Computing and Networking (MobiCom), pp. 134–146, San Diego, CA,
2003.
[19] Jian Ma, Qian Zhang, Chen Qian, and Lionel M. Ni. Energy-Efficient Opportunistic Topology
Control in Wireless Sensor Networks. In Proceedings of the 1st International MobiSys Workshop
on Mobile Opportunistic Networking (MobiOpp), pp. 33–38, New York, NY, 2007. ACM.
[20] Ilias Leontiadis and Cecilia Mascolo. GeOpps: Geographical Opportunistic Routing for
Vehicular Networks. In Proceedings of the IEEE International Symposium on a World of
Wireless Mobile and Multimedia Networks (WoWMoM), pp. 1–6, Helsinki, Finland, June 2007.
[21] BoZidar Radunović, Christos Gkantsidis, Peter Key, and Pablo Rodriguez. An Optimization
Framework for Opportunistic Multipath Routing in Wireless Mesh Networks. In Proceedings
of the 27th IEEE International Conference on Computer Communication (INFOCOM), pp.
2252–2260, April 2008.
[22] Dimitrios Koutsonikolas, Chih-Chun Wang, and Y. Charlie Hu. CCACK: Efficient Network
Coding Based Opportunistic Routing through Cumulative Coded Acknowledgments. In
Proceedings of the 29th IEEE International Conference on Computer Communication
(INFOCOM), pp. 2919–2927, Piscataway, NJ, 2010. IEEE Press.
[23] Rahul C. Shah, Sven Wiethölter, Adam Wolisz, and Jan M. Rabaey. When Does Opportunistic
Routing Make Sense? In Proceedings of the 3rd Annual IEEE International Conference on
Pervasive Computing and Communications (PerCom), pp. 350–356, March 2005.
 Opportunistic Routing in Mobile Networks  297

[24] Jonghyun Kim and Stephan Bohacek. A Comparison of Opportunistic and Deterministic
Forwarding in Mobile Multihop Wireless Networks. In Proceedings of the 1st International
MobiSys Workshop on Mobile Opportunistic Networking (MobiOpp), pp. 9–16, New York,
NY, 2007. ACM.
[25] Kai Zeng, Wenjing Lou, and Hongqiang Zhai. On End-to-End Throughput of Opportunistic
Routing in Multirate and Multihop Wireless Networks. In Proceedings of the 27th IEEE
International Conference on Computer Communication (INFOCOM), pp. 816–824, April
2008.
[26] Llorenc Cerdà-Alabern, Vicent Pla, and Amir Darehshoorzadeh. On the Performance Modeling
of Opportunistic Routing. In Proceedings of the 2nd International Workshop on Mobile
Opportunistic Networking (MobiOpp), pp. 15–21, New York, NY, 2010. ACM.
[27] Mei-Hsuan Lu, Peter Steenkiste, and Tsuhan Chen. Video Transmission over Wireless Multihop
Networks Using Opportunistic Routing. In Proceedings of the 16th IEEE International Packet
Video Workshop (PV), pp. 52–61, November 2007.
[28] Fan Wu, Tingting Chen, Sheng Zhong, Li Erran Li, and Yang Richard Yang. Incentive-Compatible
Opportunistic Routing for Wireless Networks. In Proceedings of the 14th ACM International
Conference on Mobile Computing and Networking (MobiCom), pp. 303–314, New York, NY,
2008. ACM.
[29] J. J. Garcia-Luna-Aceves and Shree Murthy. A Path-Finding Algorithm for Loop-Free Routing.
IEEE/ACM Transactions on Networking, 5:148160, 1997.
[30] Jochen Behrens and J. J. Garcia-Luna-Aceves. Distributed, Scalable Routing Based on Link-State
Vectors. In Proceedings of ACM SIGCOMM, pp. 136–147, 1994.
[31] Kirill Levchenko, Geoffrey M. Voelker, Ramamohan Paturi, and Stefan Savage. XL: An Efficient
Network Routing Algorithm. In Proceedings of ACM SIGCOMM, pp. 15–26, 2008.
[32] Shree Murthy and J. J. Garcia-Luna-Aceves. An Efficient Routing Protocol for Wireless Networks.
Mobile Networks and Applications, 1(2):183–197, 1996.
[33] J. J. Garcia-Luna-Aceves and Marcelo Spohn. Source-Tree Routing in Wireless Networks. In
Proceedings of the 7th Annual International Conference on Network Protocols (ICNP’99), pp.
273–282, Toronto, Canada, October 1999.
[34] Xin Yu. Distributed Cache Updating for the Dynamic Source Routing Protocol. IEEE
Transactions on Mobile Computing, 5(6):609–626, 2006.
[35] Yih-Chun Hu and David B. Johnson. Implicit Source Routes for On-Demand Ad Hoc Network
Routing. In Proceedings of the 3rd ACM International Symposium on Mobile Ad Hoc
Networking and Computing (MobiHoc ’01), pp. 1–10, Long Beach, CA, October 2001.
[36] Bin Hu and Hamid Gharavi. DSR-Based Directional Routing Protocol for Ad Hoc Networks. In
Proceedings of the 2007 IEEE Conference on Global Telecommunications (GLOBECOM), pp.
4936–4940, Washington, DC, November 2007.
[37] Douglas West. Introduction to Graph Theory (2nd Edition). Prentice Hall, Upper Saddle River,
NJ, August 2000.
[38] Zehua Wang, Yuanzhu Chen, and Cheng Li. CORMAN: A Novel Cooperative Opportunistic
Routing Scheme in Mobile Ad Hoc Networks. IEEE Journal on Selected Areas in Communications,
to appear in 2012.
[39] Zehua Wang, Cheng Li, and Yuanzhu Chen. PSR: Proactive Source Routing in Mobile Ad
Hoc Networks. In Proceedings of the 2011 IEEE Conference on Global Telecommunications
(GLOBECOM), Houston, TX, December 2011.
[40] Zehua Wang, Yuanzhu Chen, and Cheng Li. A New Loop-Free Proactive Source Routing Scheme
for Opportunistic Data Forwarding in Wireless Networks. IEEE Communications Letters,
15(11):1184–1186, 2011.
[41] Zehua Wang, Yuanzhu Chen, and Cheng Li. PSR: A Lightweight Proactive Source Routing
Protocol for Mobile Ad Hoc Networks. In 2012 IEEE Transactions on Vehicular Technology,
63(2):859–868, 2014 (February), doi: 10.1109/TVT.2013.2279111.
[42] Zehua Wang, Cheng Li, and Yuanzhu Chen. Local Cooperative Retransmission in Opportunistic
Data Forwarding in Mobile Ad-hoc Networks. In 2012 IEEE International Conference on
Communications (ICC), Ottawa, ON, 2012, pp. 51–62, doi: 10.1109/ICC.2012.6364129.
298  Opportunistic Networks

[43] Charles Reis, Ratul Mahajan, Maya Rodrig, David Wetherall, and John Zahorjan. Measurement-
Based Models of Delivery and Interference in Static Wireless Networks. In Proceedings of ACM
SIGCOMM, pp. 51–62, August 2006.
[44] Mei-Hsuan Lu, Peter Steenkiste, and Tsuhan Chen. Design, Implementation and Evaluation
of an Efficient Opportunistic Retransmission Protocol. In Proceedings of the 15th Annual
International Conference on Mobile Computing and Networking (MobiCom), pp. 73–84, New
York, NY, 2009. ACM.
[45] White Paper Wild Packets Inc. Converting Signal Strength Percentage to dBm Values, November
2002.
[46] Mahesh K. Marina and Samir R. Das. Routing Performance in the Presence of Unidirectional
Links in Multihop Wireless Networks. In The Third ACM International Symposium on Mobile
Ad Hoc Networking and Computing (MobiHoc ’02), pp. 12–23, Lausanne, Switzerland, June
2002.
Chapter 14

Security in Opportunistic Networks

Mohini Singh and Anshul Verma
Banaras Hindu University, Varanasi, India

Pradeepika Verma
Indian Institute of Technology (BHU), Varanasi, India

CONTENTS

14.1 Introduction .........................................................................................................299

14.2 Characteristics of Opportunistic Networks........................................................... 300
14.3 Issues in Opportunistic Networks.......................................................................... 301
14.4 Security Issues in Opportunistic Networks............................................................ 302
14.5 Thrust Areas of Security........................................................................................ 303
14.6 Security Architecture for Opportunistic Networks.................................................305
14.7 Security Attacks and Their Defense Mechanisms................................................... 306
14.8 Conclusion............................................................................................................ 309
Acknowledgment............................................................................................................. 309
References........................................................................................................................ 309

14.1 INTRODUCTION

The opportunistic network does not require an end-to-end connection between source and
destination nodes for communication. It is the best way for providing communication in an
intermittent mobile ad hoc network. It may consist of innumerable nodes that are frequently
connected or disconnected due to the mobility of the nodes. The messages are exchanged or
forwarded by exploiting expedient contacts between nodes without relying on pre-existing
network infrastructure [1]. The delay tolerant network is a superclass of the opportunistic
network [2].
The opportunistic network works on the store-carry-and-forward mechanism [3, 4]. As
shown in Figure 14.1, the source sends a message to the destination, but there is no direct
communication link between them. Therefore, some intermediate nodes are utilized to for-
ward the message towards the destination. The intermediate nodes store and carry the mes-
sage until they find the destination node or better intermediate node to pass the message to
the destination. The routing protocols also exploit much of the users’ context information to
optimize the routing task [5–7].
The opportunistic network poses various unique features, such as high node mobility, inter-
mittent communication links, and long communication delays. Therefore, designing oppor-
tunistic forwarding protocols are very difficult and vulnerable to various security issues. The
design of opportunistic networks faces serious problems, such as how to effectively imple-
ment node authentication and access control, confidentiality, and data integrity and how
to ensure routing security, privacy protection, cooperation, and trust management. In other
words, systematic research on security solutions for opportunistic networks is still open and
challenging [8]. Security is a process or technique of ensuring the integrity, confidentiality,

299
300  Opportunistic Networks

Figure 14.1  Opportunistic network.

non-repudiation, availability, authorization, and authentication of a network system. Despite

the various challenges, many real-life applications have been developed based on opportu-
nistic networks [9].
The rest of the chapter is organized as follows. Section 14.2 describes characteristics of
opportunistic networks. Sections 14.3 and 14.4 discuss general issues and security issues
in opportunistic networks, respectively. A study of main thrust areas of security as well as
research works that have been done in these areas are presented in Section 14.5. Section 14.6
describes the general architecture of the security in opportunistic networks. Section 14.7
presents various security attacks in opportunistic networks and their defense mechanisms.
Finally, we conclude the chapter and give some future research directions.

14.2 CHARACTERISTICS OF OPPORTUNISTIC NETWORKS

Many developing and potential communication circumstances do not accept the web’s under-
lying presumptions. In a lack of the end-to-end path between source and destination, the
TCP/IP protocol does not work properly. Some drawback occurs, such as intermittent con-
nectivity, high delay, and high error rate. However, the opportunistic network is suitable for
an environment that is characterized by intermittent connectivity, high delay, and error rate.

Mobility: Mainly, an opportunistic network is formed by handheld portable mobile devices/

nodes carried by users. Therefore, nodes are highly mobile and frequently change their
locations that are the main reasons for intermittent connectivity. However, mobility is
treated as an advantage rather than a drawback because the mobility of nodes creates
an opportunity to meet with the destination or better forwarding nodes.
 Security in Opportunistic Networks  301

Intermittent connectivity: Opportunistic networks improve network communication

where the connectivity of the network is intermittent/irregular. An opportunistic net-
work compensates for intermittent connectivity via the “store-carry-forward” tech-
niques. Military applications in opportunistic network regions are allowing critical
information retrieval in mobile battlefield circumstances using only intermittent con-
nectivity network communication. Intermittent connectivity might be scheduled or
opportunistic.
Long delay: The dynamic nature of opportunistic networks affects the connectivity
behavior in topology. Packet loss is one of the challenges for the wireless network. But
opportunistic networks use packet loss challenge as an opportunity with the help of
“store-carry-forward” techniques. In the opportunistic network, nodes store the mes-
sage until the better forwarding node comes into the communication range. These tech-
niques increase the delay of the network.
High error rate: The higher error rate may be less trustworthy for communication or data
transmission. A high error rate is the outcome of intermittent connectivity. The degree
of errors encountered during data transmission over a communication or network con-
nection is responsible for corrections in bit errors. So, it requires more bit processing or
packet retransmission as well as increases the traffic in the network. On the other hand,
the link error rate requires low retransmission. Hop-to-hop transmission reduces the
number of packets retransmitted in comparison to Internet-type transmission.

14.3 ISSUES IN OPPORTUNISTIC NETWORKS

Bandwidth: An opportunistic network has a limited bandwidth in comparison to a wired

or traditional wireless network. The opportunistic network has a low capacity as com-
pared to infrastructure networks. The opportunistic network has deficient interference
conditions and multiple accesses and fading effects.
Connectivity: Due to the limitations of the opportunistic network, connectivity is a chal-
lenge. Short-range communication and mobility of nodes decrease the connection
between the nodes in the opportunistic network. Low connectivity might affect the
topology as well as end-to-end service for forwarding data in the opportunistic net-
work. High connectivity increases the efficiency of the performance.
Congestion: An opportunistic network supports intermittent connectivity. It overcomes
delays as well as communication disruptions, which are drawbacks of a traditional
network. Data transmission is allowed between the source and the destination, where
the end-to-end path does not exist. In the opportunistic network, nodes’ connectiv-
ity is limited, but some nodes are well connected in the network. Hence, these nodes’
connections are exploited by most of the ongoing transmissions; as a consequence,
these connections become more vulnerable to congestion. It also increases the unusable
disconnection and low delivery rate. Congestion control [10, 11] is a more challenging
issue in the opportunistic network because of intermittent connectivity. The oppor-
tunistic network mainly targets the management of buffer for congestion control by
eliminating unrequired data from the buffer.
Routing: An opportunistic network observes issues related to end-to-end services, where
the pre-existing path does not exist. The dynamic behavior of nodes increases the chal-
lenge for route establishment. It is necessary to overcome main issues such as mobility,
dynamic topology, and routing table. So, an opportunistic network is forwarding data
using the “store-carry-forward” technique [8–10].
302  Opportunistic Networks

a. Mobility: Mobility is a reason for the dynamic behavior of the network [12, 13].
Due to mobility, nodes frequently change their locations in the network and might
increase the coverage area of the network.
b. Topology: Due to the frequent mobility of nodes, the topology between nodes is
also dynamic. Sometimes a temporary topology is established between the nodes,
but it is not stable due to the nodes’ mobility.
c. Routing table: In the opportunistic network, unlike traditional wireless networks,
pre-existing path information is not saved in routing tables because nodes are
highly mobile; they change their positions frequently. If a node changes its position,
that affects the routing table. The routing table frequently changes its records, and
updating routing tables at each node frequently generates extra overhead.

Security: An opportunistic network observes security issues that are concerned with end-
to-end path services as well as where the end-to-end path may not exist for data for-
warding. It is necessary to have some authorized, authentication, privacy, and access
control mechanism to handle the security-related issues in the network. Various objec-
tives have been created for the security component of the opportunistic network. The
first objective is immediately avoiding unapproved applications from having their infor-
mation brought through the opportunistic network. The second objective is avoiding
unapproved applications from claiming control over the opportunistic network. The
third objective is to instantly dispose of groups that are harmful or inappropriate for
the network. In conclusion, the objective is to recognize expeditiously and de-approve
bargained elements.

14.4 SECURITY ISSUES IN OPPORTUNISTIC NETWORKS

Mostly, security of the network techniques is a challenge to collective message integrity and
authorized identities of users. These techniques do not challenge routers’ authorization,
which processes the information for end-to-end services. This segment examines a portion of
the issues that occur in the security of an opportunistic network.

Key management: A significant open issue in opportunistic network security is the absence
of a delay-tolerant technique for key administration [14]. At this phase, we are just
truly realizing how to utilize existing plans, which at last require an online status testing
service or key dissemination service, which isn’t viable in a high delay or exceptionally
interrupting condition. The main, for the most part, appropriate plans we have right
now are virtually identical to shared confidential facts or, in all likelihood, irreversible
public key (or authentication-based) plans. These are the regions where more research
work could create fascinating outcomes. The key management is based on the typically
traditional cryptography. It does not sustain router authentication.
Replay management: Handling replay is one of the significant issues in the opportunistic
network. It became a challenge to discourse replay issues in the opportunistic network
because of the intermitted connectivity in the network. In major networking circum-
stances, we either try to remove or else drastically minimize the possibility of a message
being replayed. The key management based on the handling replays depends mostly on
the message possibly being replayed. In some opportunistic network circumstances, this
determination also is mostly the case, as replaying (for example, authorized or authen-
ticated) data can be a reasonably straightforward method to consume rare network
resources. Delay elements in the opportunistic network also make handling the replay
 Security in Opportunistic Networks  303

a difficult task. Therefore, replay management remains an open security issue in the
opportunistic network.
Security of routing protocols: A major issue in the opportunistic network is routing proto-
col security. On the other hand, if any opportunistic network routing protocol was sup-
posed to use either Licklider Transmission Protocol (LTP) [15–17], or Bundle Protocol,
it could utilize their available security features. The mechanism of security is established
for the block of metadata. This mechanism has been used for another block of non-
payload data. For some issues, it might provide solutions, but most of the issues are still
unaddressed.
Traffic investigation: A general traffic investigation assurance plot is not a reasonable
objective for the opportunistic network, and there has been no requirement of a non-
exclusive way to deal with this issue. For some disturbance lenient systems, concealing
traffic might be a significant security necessity. Furthermore, the leading issue is the
degree to which there is a genuine requirement for a nonexclusive plan for security
against traffic investigation. The subsequent issue is how to characterize such a plan
to be postponement- and interruption-lenient and which additionally doesn’t devour
an excessive number of assets. At last, traffic investigation insurance might be left as a
nearby issue for the fundamental system layers.
Multicast security: In the opportunistic network, there is a lack of techniques that define
which nodes register for anycast or multicast end-point services. Therefore, there is
a need for techniques that will clearly define which node is registered for anycast or
multicast for a particular service. For example, the signup for a mailing list is similar to
multicast end-point registration.
Evaluation of performance: The security arrangement for the opportunistic network will
expend extra bandwidth utilization cost and computation cost for the network’s nodes.
Also, if more than one security management techniques are utilized, greater bandwidth
amount and computation power will probably be utilized. Therefore, security tech-
niques should be developed according to the need of the applications, bandwidth, and
computation power available for the security features; it should not generate extra
overhead on the network.

14.5 THRUST AREAS OF SECURITY

Security is an important issue in opportunistic networks; therefore, several research works

have been done to resolve various aspects of security. Here we discuss research works that
have been done related to bundle security protocols, key and trust management, security
initialization, fragment authentication, etc.

Bundle security protocols: Bundle security protocols [18] provide a guarantee for hop-
by-hop authentication, end-to-end authentication, and end-to-end confidentiality of
messages, and they are described in the RFC 6257 standard. The Bundle Protocol speci-
fication [19] has four extension blocks that are added to a message, namely the Bundle
Authentication Block (BAB), Payload Integrity Block (PIB), Payload Confidentiality
Block (PCB), and the Extension Security Block (ESB). However, the protocol does not
include a key management module and assumes that all nodes must have some provision
to perform cryptographic operations. BAB is responsible for the hop-to-hop authentica-
tion of a message between two adjacent nodes. PIB ensures end-to-end authentication
of a message between the source and destination. PCB ensures end-to-end confidential-
ity by encrypting messages between the source and destination. Whereas, ESB protects
304  Opportunistic Networks

the metadata portion (non-payload part) of a message. The standard also defines the
way of using cryptographic approaches to process the security blocks on the intermedi-
ate nodes or final destination node. The standard also has the provision of a security-
zone, security-source, and security-destination, which divides the security control over
multiple network organizations. This also provides security to the devices that do not
support cryptographic operations.
Key management: Secure key distribution is necessary for a network to ensure message
integrity, authenticity, and confidentiality. In the traditional Internet, secure key distribu-
tion is implemented by using online servers. Whereas, online servers cannot be assumed
in opportunistic networks for key distribution and certificate verification [20]. To solve
this problem, Identity Based Cryptography (IBC) [21] was proposed in [20] for secure
key distribution in intermittently connected networks. Time-based keys [22] were used
for controlling the rights of compromised nodes and malicious users, in which a signing
key is valid for a short period. To check the applicability of the IBC approach, a com-
parative study between IBC and traditional PKI approaches was performed in [23] over
message integrity, authenticity, and end-to-end confidentiality. It was concluded that
IBC does not provide any advantage in opportunistic networks over the traditional PKI
approach in terms of message integrity and authenticity. However, it was shown that
IBC performs better than the PKI approach in terms of end-to-end confidentiality. A
security architecture was proposed in [24] that introduces physical and cryptographic
approaches to protect Internet kiosks. Based on a similar concept, Internet kiosks were
used in [25] as PKGs to generate users’ private keys. However, this approach does not
guarantee user authentication.
Security initialization: Bootstrapping security in opportunistic networks is a challenging
issue. An existing mobile network infrastructure-based approach was proposed in [23]
that is assumed to be a widely available authentication scheme. A social context infor-
mation-based approach was presented in [26] that uses social context information of
nodes such as knowledge of present and past affiliation or encounter history of nodes
to create an initial security context between two nodes that do not have any security
relationship. They used the distribution of keys within some affiliated entities that are
common for both nodes, and through those keys, they established secure communica-
tion between these two nodes.
Trust management: The main goal of trust management is preventing malicious nodes
from conducting Sybil and black hole attacks. In the Sybil attack [27], a malicious
node creates multiple identities in the network to get a larger influence in the net-
work. The black hole attack [28] is a type of denial of service (DoS) attack in which
messages are dropped by malicious nodes or other compromised nodes, malicious
nodes may pretend to be a destination and may attract messages. An approach was
developed in [29] to initializing the trust by building certificate chains (such as the
Pretty Good Privacy system) that assume unconditional transitivity of trust along the
chain path. In [30], social trust is used to detect Sybil attacks. Two types of social
trusts (Explicit and Implicit social trusts) were used to detect Sybil attacks. The result
of this approach is used to create an integrated framework for the secure content
dissemination known as PodNetSec [31], which is built on top of PodNet [32]. To
prevent black hole attacks, a reputation-based approach was proposed in [33] to pre-
dict the compatibility of an encountered node to deliver a message to a destination.
In their extended work [34], the authors have applied it on PRoPHET to investigate
its effectiveness against black hole attacks. A data-centric trust-based framework was
presented in [35] to prevent black hole attacks, whose validity is inferred based on
the Dempster-Shafer theory [36].
 Security in Opportunistic Networks  305

Fragment authentication: A secure message fragmentation scheme is proposed in [37] that

is named toilet paper, in which checkpoints are inserted into fragments by using a
cryptographic hash. A signature is also inserted in the message along with hashes that
help any intermediate node to authenticate the fragment by using them. An enhanced
version of this approach is proposed in [23]. A special authentication function is pro-
posed in [37] for fragments authentication; however, how to construct such an efficient
function is not described.

14.6 SECURITY ARCHITECTURE FOR OPPORTUNISTIC NETWORKS

The opportunistic network utilizes interactions of intermediate nodes to forward messages

towards a destination. Intermediate nodes may be unreliable, and that raises various seri-
ous issues related to authentication, authorization, security, and privacy. To overcome these
issues, a general security architecture is proposed in [8] (see Figure 14.2) that includes five
modules: authentication, secure routing, access control, trust management and cooperation,
and application/user-specific privacy protection. However, all modules may not be necessary
for all opportunistic networks, and one or more modules may be eliminated depending upon
user demands, network requirements, or threat scenarios.
First, the authentication security module verifies the identity of the nodes to prevent unau-
thorized nodes from joining the opportunistic networks. Second, the access control module
verifies whether a particular node or user is authorized to access a network resource or to
perform an operation. Third, the secure routing module ensures the message contents’ con-
fidentiality and avoids message tampering as well as delivers the message to the destination
eventually. Fourth, the trust management and cooperation module monitors the behavior of
the intermediate nodes to detect any malicious activity as well as implements a trust evalu-
ation mechanism to promote nodes’ cooperation and prevent selfish behavior of the nodes.
Finally, the application/user-specific privacy protection module provides customization of
privacy features to the nodes or applications according to their needs. This means a node or
application may express or hide some information related to them selectively.

Figure 14.2  Security Architecture for Opportunistic Networks.

306  Opportunistic Networks

14.7 SECURITY ATTACKS AND THEIR DEFENSE MECHANISMS

In opportunistic networks, attacks can be classified into two classes: internal and external
attacks [38–41]. Internal attacks occur due to the internal nodes that are part of the network
and therefore are more harmful for the network because internal nodes have access rights
over various network resources. Whereas, external attacks occur due to the external nodes
those are not part of the network therefore are less harmful in comparison internal attacks
because they do not have access rights over the network resources. Opportunistic networks
are vulnerable to the various attacks such as wormhole attacks, black hole attacks, selfish
attacks, Sybil attacks, and selective dropping attacks. These attacks and their defense mecha-
nisms are discussed as follows.

Wormhole attacks: In wormhole attacks [42], malicious nodes pretend to be the source
node of some packets in the network; those are actually sent by another node. Malicious
nodes may have shortest path for some destinations and therefore may attract traf-
fic or other nodes. The packet leash technique was proposed in [43] to protect from
wormhole attacks in which geographical leashes and temporal leashes information can
be added to packets to protect them from the attacks. This technique requires secure
and tight time synchronization. Geographical leashes restrict the distance between the
source and destination of a packet. Whereas, temporal leashes restrict the lifetime of a
packet, and as a consequence, it restricts the maximum distance travelled by a packet.
However, this kind of technique needs secure and tight time synchronization. An ad hoc
on-demand distance vector (AODV) protocol-based wormhole attack detection and
isolation technique was proposed in [44]. The source node sends a route request to the
destination and in response receives all available routes with the number of hops. These
routes are used to detect malicious nodes by applying route redundancy and routes
aggregation and calculating the round-trip time (RTT) for all listed routes. However,
this approach is not feasible for the opportunistic networks because end-to-end connec-
tions between source and destination may not be present, and it is very hard to find out
more than one route to the destination.
Black hole attacks: In black hole attacks, malicious nodes discard some or all of the
received packets and advertise that they have route information of the most demand-
ing destinations on the network. An “encounter ticket” (ET)-based defense technique
was proposed in [45] against a black hole attack in which ET represents evidence of
the encounters of nodes. However, this technique does not work in all cases, and it
detects an attacker only when claiming non-existent encounters and does not handle
packet discarding due to black hole attack. A watchdog-based approach was pro-
posed in [34, 46] to detect black hole attacks in which a trust/reputation is estimated
for each node in the network to detect whether it is a malicious node or not. Trust/
reputation is estimated for each node by other nodes based on its message forwarding
history. Say a node A forwards a message to node B to deliver the message to node
C. Node C sends a positive feedback message (PFM) to node A on successfully recep-
tion of the message from node B; otherwise, node A registers node B as a suspicious
node until PFM arrives. In this way, the trust/reputation system is built. However,
PFM is sent using epidemic routing that generates network overhead. An approach
based on maintaining secure packets exchanging history between nodes is proposed
in [47]. When two nodes encounter, they record the number of exchanged packets
after encryption using a private key to prevent black hole attacks. However, initial-
izing keys to the nodes manually or using a key distribution scheme will be difficult
in opportunistic networks.
 Security in Opportunistic Networks  307

Dropping and selective dropping attacks: In a normal situation, packets are dropped
according to the predefine policies of the network, such as resource limitation or
expiration of time to live (TTL). A packet weight-based dropping policy and mecha-
nism is proposed in [48] where weight is estimated using inter-contact time between
nodes. However, it is very difficult to detect dropping or selective dropping attacks
because malicious nodes drop few or all received packets, and source and destination
nodes do not know where and when dropping takes place. Authenticated end-to-end
or link-to-link acknowledgment mechanisms can be used for detecting packet drop-
ping attacks [49, 50]. A network coding-based mechanism is proposed in [51] to
assess the effect of a packet selective dropping attack. A destination node calculates
delivery ratio and sends it to the source node. The source node starts dynamically
adjusting the redundancy factor to cover the degradation in the delivery ratio caused
by the attack. The effect of packet dropping on routing performance is evaluated by
theoretical analysis as well as simulation. In [52], a packet dropping attack detection
mechanism is proposed in which intermediate nodes send an acknowledgment to
the source node on safe reception of a packet. Source node constructs a Markle tree
on the basis of received acknowledgments and then compares the value of the tree
root with precalculated value. If both the values are equal, then no packet dropping
occurred on this path; otherwise, there is packet dropping. This mechanism can detect
the path with the malicious node and can suggest alternative paths for retransmis-
sion. However, this mechanism generates network overhead and does not detect the
exact malicious node on the path. A cooperative participation-based packet dropping
detection mechanism is proposed in [53] in which cooperative participation occurs
at the network-bootstrapping phase. Alternative routes are chosen to avoid the mali-
cious or non-trusted path. This mechanism generates network overhead. A data prov-
enance-based packet dropping attack detection technique is proposed in [54]. The
watermarking-based secure provenance transmission mechanism and the inter-packet
timing are used to identify malicious nodes. The technique performs three main tasks:
detecting lost packets, identifying the presence of the attacks, and identifying and
isolating the malicious paths. However, the technique does not identify the exact mali-
cious node on the path.
Watchdog and pathrater techniques are used in [55] to improve throughput. The watch-
dog technique is used to detect the malicious node by monitoring the packets’ trans-
mission pattern. Whereas, the pathrater technique is used to predict the best path by
using the information from the watchdog and the link reliability data. The watchdog
technique is not efficient in case of ambiguous collisions, receiver collisions, false misbe-
havior, and limited transmission power. An ExWatchdog technique is proposed in [56]
to overcome the drawbacks of the watchdog technique and to enhance the malicious
node detection process. ExWatchdog is capable of detecting those malicious nodes who
partition the network by falsely declaring other nodes as malicious. A reputation-based
technique for detecting packet dropping attacks is presented in [57], in which direct and
indirect observation information is used to estimate full reputation weight. Nodes are
included or excluded from the network on the basis of reputation weights. Moreover,
historical reputation and fuzzy logic are used to improve the performance of the fault
tolerance. A packet-dropping-attack detection and malicious node detection algorithm
is presented in [58]. The algorithm uses an indicative field in the packet header that is
used to investigate whether a node received the complete original number of packets
from the previous node or not. An attack detection mechanism is proposed in [59] to
detect a special type of packet dropping attack where malicious nodes inject fake pack-
ets in place of dropped packets. The creation time of the packets are used to predict
308  Opportunistic Networks

whether they are original or fake. The results show the mechanism achieves high accu-
racy and detection rate.
Selfish and Sybil attacks: Selfish nodes use network resources and services but deny to
offer services for other nodes due to some constraints, such as, selfish nodes do not
forward packets due to low battery or buffer overflow. Defense techniques to protect
from selfish attack are classified into two classes: barter based and credit based. A
barter-based technique to encourage selfish nodes to cooperate is proposed in [60]. The
technique has two major parts: a reputation system and a virtual payment or rewarding
system. When two nodes encounter, they exchange summaries of the messages currently
in their buffer. Thereafter, they decide which messages have to be exchanged and in
which order. Messages are exchanged one by one from both sides. If any node cheats,
the transmission is canceled. After each successful message transmission, the nodes
receive a score that is accumulated to calculate their total score at the end. However,
this technique assumes both that the encountered nodes have the same number of mes-
sages to be exchanged, and that the connection time will be enough to exchange all the
decided messages that are not practical in real scenarios. A credit-based incentive tech-
nique named MobiCent is proposed in [61], in which each node pays the relay nodes
for forwarding messages. The payment scheme uses two algorithms: first is a payment
set selection algorithm that decides which relays to be paid, and second is a payment
calculation algorithm that decides how much has to be paid to each selected relay. Due
to rewards, each relay node cooperates in the forwarding of messages without hesita-
tion unless the reward is not sufficient. This technique will not effectively work if the
network has selfish nodes in the majority.
In a Sybil attack, malicious nodes drop received packets as well as create fake identities.
Therefore, it is very difficult to detect an actual malicious node because it uses different
identities to communicate with neighbors. The definition and taxonomy of the Sybil
attack is described in [62]. Various defense techniques are classified as old and new
techniques. For example, the resource testing is described as an old technique, and radio
resource testing, verification of key sets for random key predistribution, registration,
position verification, and code attestation are shown as new defense techniques for
Sybil attacks. A reputation-based defense technique to detect Sybil attacks is described
in [63], in which explicit and implicit social trust establishment mechanisms are used.
The trust is established between nodes by the contact quality between nodes and the
trustworthiness of the nodes’ opinions.
Anti-localization techniques: A node location can be predicted by tracking the path and
movement of the node. The node location information can be misused by malicious
nodes; thus, it is a serious concern in opportunistic networks. ALAR [64] is one of the
techniques used to implement location privacy. ALAR divides a message into many
encrypted segments and sends them to destination through different paths. The decryp-
tion key is stored in the last segment; therefore, a node cannot interpret the message
contents unless it receives all the segments. The technique maximizes the sender’s loca-
tion privacy as well as message delivery. However, delivery ratio and delivery latency
are degraded due to overhead of setting number of segments and multiple paths. In
[65], nodes of the network are divided into groups; each node must be a member of
a group. Each node is assigned its public/private key pair as well as its group’s public/
private key pair. Each node also maintains a copy of the public keys of all other nodes
and groups. The source node divides a message into multiple parts and encrypts these
parts using the destination node’s public key. The source inserts the destination address
into each encrypted message part and sends all the parts through multiple paths. Each
node checks whether it is the destination of the message; if it is, it decrypts the message,
 Security in Opportunistic Networks  309

otherwise it is forwarded to the neighbors. However, this technique is vulnerable to

Sybil attacks. The application of k-anonymity for location-based services (LBS) and its
recent advancements are reviewed in [66]. The applicability of k-anonymity for LBS is
mainly classified on the basis of architecture, algorithms for anonymization, and the
types of k-anonymity (according to the different query processing techniques).

14.8 CONCLUSION

The aim of this chapter was to discuss main security issues in opportunistic networks, and the
research works have been done to address these issues. The chapter described the opportu-
nistic network and its characteristics. The chapter discussed general issues as well as security
issues of opportunistic networks. Moreover, various thrust areas of security were discussed
along with the research works that have been done to address issues of these thrust areas. A
general security architecture for opportunistic networks was also described. Finally, various
security attacks and their defense techniques were discussed. As a new networking paradigm,
opportunistic networks have very potential real-life applications in diverse domains such as
wildlife monitoring, underwater networks, mobile social networking, pervasive computing,
Internet communication in rural areas, and intelligent transportation. Therefore, security,
privacy, and trust issues that are considered serious challenges for opportunistic networks
must be addressed.
For future research, developing cooperative incentive mechanisms and fairness-aware
incentive schemes for trust management may be good research directions. Moreover, the
data-centric trust mechanism rather than the entity trust mechanism should be deeply stud-
ied in opportunistic networks. Privacy preservation is also a main concern for opportunistic
networks that should be further investigated. Mostly, the correctness of security and trust
protocols are validated through the simulators; that is not sufficient. Therefore, formal verifi-
cation of security and trust protocols by using formal methods will be an emerging research
direction in opportunistic networks.

ACKNOWLEDGMENT

This chapter is a part of the research work funded by “Seed Grant to Faculty Members under
IoE Scheme (under Dev. Scheme No. 6031)”1 and “DST-Science and Engineering Research
Board (SERB), Government of India (File no. PDF/2020/001646)”2.

NOTES

1 Awarded to Anshul Verma at Banaras Hindu University, Varanasi, India.

2 Awarded to Pradeepika Verma at Indian Institute of Technology (BHU), Varanasi, India.

REFERENCES

[1] Wahid, A., Kumar, G., & Ahmad, K. (2014). Opportunistic networks: Opportunity versus chal-
lenges-survey. In National Conference on Information Security Challenges (pp. 14–24).
[2] Boldrini, C., Conti, M., & Passarella, A. (2008). Autonomic behaviour of opportunistic network
routing. International Journal of Autonomous and Adaptive Communications Systems, 1(1),
122–147.
310  Opportunistic Networks

[3] Ciobanu, R. I., & Dobre, C. (2013). Opportunistic networks: A taxonomy of data dissemina-
tion techniques. International Journal of Virtual Communities and Social Networking (IJVCSN),
5(2), 11–26.
[4] Bharamagoudar, S. R., & Saboji, S. V. (2017, December). Routing in opportunistic networks:
taxonomy, survey. In 2017 International Conference on Electrical, Electronics, Communication,
Computer, and Optimization Techniques (ICEECCOT) (pp. 300–305). IEEE.
[5] Verma, A., & Srivastava, A. (2011). Integrated routing protocol for opportunistic networks.
International Journal of Advance Computer Science and Application, 2(3), 85–91.
[6] Verma, A., Pattanaik, K. K., & Ingavale, A. (2013). Context-based routing protocols for oppnets.
In Routing in Opportunistic Networks (pp. 69–97). Springer, New York, NY.
[7] Verma, A., & Pattanaik, K. K. (2017). Routing protocols in opportunistic networks. Opportunistic
Networking: Vehicular, D2D and Cognitive Radio Networks, 125–166.
[8] Wu, Y., Zhao, Y., Riguidel, M., Wang, G., & Yi, P. (2015). Security and trust management in
opportunistic networks: a survey. Security and Communication Networks, 8(9), 1812–1827.
[9] Verma, A., Singh, M., Pattanaik, K. K., & Singh, B. K. (2018). Future Networks Inspired by
Opportunistic Networks. Opportunistic Networks: Mobility Models, Protocols, Security, and
Privacy, 104, 229–246.
[10] Grundy, A. (2012). Congestion control framework for delay-tolerant communications (Doctoral
dissertation, University of Nottingham).
[11] Coe, E., & Raghavendra, C. (2010, March). Token based congestion control for DTNs. In 2010
IEEE Aerospace Conference (pp. 1–7). IEEE.
[12] Fathima, M., Ahmad, K., & Fathima, A. (2017). Mobility models and routing protocols in oppor-
tunistic networks: a survey. International Journal of Advanced Research in Computer Science,
8(5), 354–371.
[13] Mota, V. F., Cunha, F. D., Macedo, D. F., Nogueira, J. M., & Loureiro, A. A. (2014). Protocols,
mobility models and tools in opportunistic networks: a survey. Computer Communications, 48,
5–19.
[14] Sahadevaiah, K., & Reddy, P. P. (2011). Impact of security attacks on a new security protocol for
mobile ad hoc networks. Network Protocols and Algorithms, 3(4), 122–140.
[15] Ramadas, M., Burleigh, S., & Farrell, S. (2008). Licklider transmission protocol-specification.
IETF Request for Comments RFC, 5326.
[16] Burleigh, S., Ramadas, M., & Farrell, S. (2008). Licklider transmission protocol-motivation.
IETF Request for Comments RFC, 5325.
[17] Wang, R., Burleigh, S. C., Parikh, P., Lin, C. J., & Sun, B. (2010). Licklider transmission protocol
(LTP)-based DTN for cislunar communications. IEEE/ACM Transactions on Networking, 19(2),
359–368.
[18] Symington, S., Farrell, S., Weiss, H., & Lovell, P. (2007). Bundle security protocol specification.
Work Progress, October, 28.
[19] Scott, K., & Burleigh, S. (2007). Bundle protocol specification, https://tools.ietf.org/html/rfc5050.
[20] Seth, A., & Keshav, S. (2005, November). Practical security for disconnected nodes. In 1st IEEE
ICNP Workshop on Secure Network Protocols, 2005 (NPSec). (pp. 31–36). IEEE.
[21] Boneh, D., & Franklin, M. (2001, August). Identity-based encryption from the Weil pairing. In
Annual International Cryptology Conference (pp. 213–229). Springer, Berlin, Heidelberg.
[22] Fall, K. (2003). Identity based cryptosystem for secure delay tolerant networking. Manuscript.
[23] Asokan, N., Kostiainen, K., Ginzboorg, P., Ott, J., & Luo, C. (2007). Towards securing disrup-
tion-tolerant networking. Nokia Research Center, Tech. Rep. NRC-TR-2007-007.
[24] Ur Rahman, S., Hengartner, U., Ismail, U., & Keshav, S. (2008, August). Practical security for
rural Internet kiosks. In Proceedings of the Second ACM SIGCOMM Workshop on Networked
Systems for Developing Regions (pp. 13–18).
[25] Ma, D., & Tsudik, G. (2010). Security and privacy in emerging wireless networks. IEEE Wireless
Communications, 17(5), 12–21.
[26] El Defrawy, K., Solis, J., & Tsudik, G. (2009, July). Leveraging social contacts for message con-
fidentiality in delay tolerant networks. In 2009 33rd Annual IEEE International Computer
Software and Applications Conference (Vol. 1, pp. 271–279). IEEE.
 Security in Opportunistic Networks  311

[27] Douceur, J. R. (2002, March). The Sybil attack. In International Workshop on Peer-to-Peer
Systems (pp. 251–260). Springer, Berlin, Heidelberg.
[28] Raymond, D. R., & Midkiff, S. F. (2008). Denial-of-service in wireless sensor networks: Attacks
and defenses. IEEE Pervasive Computing, 7(1), 74–81.
[29] Capkun, S., Buttyan, L., & Hubaux, J. P. (2003). Self-organized public-key management for
mobile ad hoc networks. IEEE Transactions on Mobile Computing, 2(1), 52–64.
[30] Trifunovic, S., Legendre, F., & Anastasiades, C. (2010, March). Social trust in opportunistic net-
works. In 2010 INFOCOM IEEE Conference on Computer Communications Workshops (pp.
1–6). IEEE.
[31] Trifunovic, S., Anastasiades, C., Distl, B., & Legendre, F. (2010, January). PodNetSec: Secure
opportunistic content dissemination. In Proceedings of ACM Mobisys (p. 1).
[32] May, M., Karlsson, G., Helgason, O., & Lenders, V. (2007, October). A system architecture
for delay-tolerant content distribution. In IEEE Conference on Wireless Rural and Emergency
Communications (WreCom).
[33] Li, N., & Das, S. K. (2010, February). RADON: reputation-assisted data forwarding in opportu-
nistic networks. In Proceedings of the Second International Workshop on Mobile Opportunistic
Networking (pp. 8–14).
[34] Li, N., & Das, S. K. (2013). A trust-based framework for data forwarding in opportunistic net-
works. Ad Hoc Networks, 11(4), 1497–1509.
[35] Raya, M., Papadimitratos, P., Gligor, V. D., & Hubaux, J. P. (2008, April). On data-centric trust
establishment in ephemeral ad hoc networks. In IEEE INFOCOM 2008-The 27th Conference
on Computer Communications (pp. 1238–1246). IEEE.
[36] Shafer, G. (1976). A Mathematical Theory of Evidence. Princeton University Press.
[37] Partridge, C. (2005, November). Authentication for fragments. In Proceedings of ACM
SIGCOMM HotNets-IV Workshop.
[38] Gupta, S., Dhurandher, S. K., Woungang, I., Kumar, A., & Obaidat, M. S. (2013, October). Trust-
based security protocol against blackhole attacks in opportunistic networks. In 2013 IEEE 9th
International Conference on Wireless and Mobile Computing, Networking and Communications
(WiMob) (pp. 724–729). IEEE.
[39] Gonçalves, M. R. P., dos Santos Moreira, E., & Martimiano, L. A. F. (2010, April). Trust man-
agement in opportunistic networks. In 2010 Ninth International Conference on Networks (pp.
209–214). IEEE.
[40] Wu, B., Chen, J., Wu, J., & Cardei, M. (2007). A survey of attacks and countermeasures in
mobile ad hoc networks. In Wireless Network Security (pp. 103–135). Springer, Boston, MA.
[41] De Fuentes, J. M., González-Manzano, L., González-Tablas, A. I., & Blasco, J. (2014). Security
models in vehicular ad-hoc networks: A survey. IETE Technical Review, 31(1), 47–64.
[42] Hu, Y. C., Perrig, A., & Johnson, D. B. (2006). Wormhole attacks in wireless networks. IEEE
Journal on Selected Areas in Communications, 24(2), 370–380.
[43] Hu, Y. C., Perrig, A., & Johnson, D. B. (2003, March). Packet leashes: a defense against worm-
hole attacks in wireless networks. In IEEE INFOCOM 2003. Twenty-second Annual Joint
Conference of the IEEE Computer and Communications Societies (IEEE Cat. No. 03CH37428)
(Vol. 3, pp. 1976–1986). IEEE.
[44] Shin, S. Y., & Halim, E. H. (2012, October). Wormhole attacks detection in MANETs using
routes redundancy and time-based hop calculation. In 2012 International Conference on ICT
Convergence (ICTC) (pp. 781–786). IEEE.
[45] Li, F., Wu, J., & Srinivasan, A. (2009, April). Thwarting blackhole attacks in disruption-tolerant
networks using encounter tickets. In IEEE INFOCOM 2009 (pp. 2428–2436). IEEE.
[46] Singh, U. P., & Chauhan, N. (2017, July). Authentication using trust framework in opportu-
nistic networks. In 2017 8th International Conference on Computing, Communication and
Networking Technologies (ICCCNT) (pp. 1–7). IEEE.
[47] Ren, Y., Chuah, M. C., Yang, J., & Chen, Y. (2010, June). Detecting blackhole attacks in dis-
ruption-tolerant networks through packet exchange recording. In 2010 IEEE International
Symposium on “A World of Wireless, Mobile and Multimedia Networks” (WoWMoM) (pp.
1–6). IEEE.
312  Opportunistic Networks

[48] Ke, M., Nenghai, Y., & Bin, L. (2010, November). A new packet dropping policy in delay toler-
ant network. In 2010 IEEE 12th International Conference on Communication Technology (pp.
377–380). IEEE.
[49] Zhang, X., Jain, A., & Perrig, A. (2008, December). Packet-dropping adversary identification for
data plane security. In Proceedings of the 2008 ACM CoNEXT Conference (pp. 1–12).
[50] Carbunar, B., Ioannidis, I., & Nita-Rotaru, C. (2004, October). JANUS: towards robust and mali-
cious resilient routing in hybrid wireless networks. In Proceedings of the 3rd ACM Workshop on
Wireless Security (pp. 11–20).
[51] Chuah, M., & Yang, P. (2009, August). Impact of selective dropping attacks on network cod-
ing performance in DTNs and a potential mitigation scheme. In 2009 Proceedings of 18th
International Conference on Computer Communications and Networks (pp. 1–6). IEEE.
[52] Baadache, A., & Belmehdi, A. (2012). Fighting against packet dropping misbehavior in multi-hop
wireless ad hoc networks. Journal of Network and Computer Applications, 35(3), 1130–1139.
[53] Sen, J., Chandra, M. G., Balamuralidhar, P., Harihara, S. G., & Reddy, H. (2007). A distributed
protocol for detection of packet dropping attack in mobile ad hoc networks. In 2007 IEEE
International Conference on Telecommunications and Malaysia International Conference on
Communications (pp. 75–80). IEEE.
[54] Sultana, S., Bertino, E., & Shehab, M. (2011, June). A provenance based mechanism to identify
malicious packet dropping adversaries in sensor networks. In 2011 31st International Conference
on Distributed Computing Systems Workshops (pp. 332–338). IEEE.
[55] Marti, S., Giuli, T. J., Lai, K., & Baker, M. (2000, August). Mitigating routing misbehavior in
mobile ad hoc networks. In Proceedings of the 6th Annual International Conference on Mobile
Computing and Networking (pp. 255–265).
[56] Nasser, N., & Chen, Y. (2007, June). Enhanced intrusion detection system for discover-
ing malicious nodes in mobile ad hoc networks. In 2007 IEEE International Conference on
Communications (pp. 1154–1159). IEEE.
[57] Song, J., & Ma, C. (2009, November). A reputation-based scheme against malicious packet
dropping for mobile ad hoc networks. In 2009 IEEE International Conference on Intelligent
Computing and Intelligent Systems (Vol. 3, pp. 113–117). IEEE.
[58] Alajeely, M., & Doss, R. (2014, September). Defense against packet dropping attacks in opportu-
nistic networks. In 2014 International Conference on Advances in Computing, Communications
and Informatics (ICACCI) (pp. 1608–1613). IEEE.
[59] Alajeely, M., Doss, R., & Mak-Hau, V. (2014, November). Packet faking attack: a novel attack
and detection mechanism in oppnets. In 2014 Tenth International Conference on Computational
Intelligence and Security (pp. 638–642). IEEE.
[60] Buttyan, L., Dora, L., Felegyhazi, M., & Vajda, I. (2007, June). Barter-based cooperation in
delay-tolerant personal wireless networks. In 2007 IEEE International Symposium on a World
of Wireless, Mobile and Multimedia Networks (pp. 1–6). IEEE.
[61] Chen, B. B., & Chan, M. C. (2010, March). Mobicent: a credit-based incentive system for disrup-
tion tolerant network. In 2010 Proceedings IEEE INFOCOM (pp. 1–9). IEEE.
[62] Newsome, J., Shi, E., Song, D., & Perrig, A. (2004, April). The sybil attack in sensor networks:
analysis & defenses. In Third International Symposium on Information Processing in Sensor
Networks, 2004. IPSN 2004 (pp. 259–268). IEEE.
[63] Trifunovic, S., Legendre, F., & Anastasiades, C. (2010, March). Social trust in opportunistic net-
works. In 2010 INFOCOM IEEE Conference on Computer Communications Workshops (pp.
1–6). IEEE.
[64] Lu, X., Hui, P., Towsley, D., Pu, J., & Xiong, Z. (2010). Anti-localization anonymous routing for
delay tolerant network. Computer Networks, 54(11), 1899–1910.
[65] Jansen, R., & Beverly, R. (2010, October). Toward anonymity in delay tolerant networks: thresh-
old pivot scheme. In 2010-Milcom 2010 Military Communications Conference (pp. 587–592).
IEEE.
[66] Zuberi, R. S., Lall, B., & Ahmad, S. N. (2012). Privacy protection through k. anonymity in loca-
tion based services. IETE Technical Review, 29(3), 196–201.
Index

A bitcoin, 188, 190

black hole attacks, 306
active messages 6–7 BlockCent, 189, 193–194, 197, 199–204
active messages, a distributed sensing blockchain, 190–191
infrastructure 18–19 blockchain-­based incentive mechanism, 190
ad hoc on-­demand distance vector (AODV), blockchain metrics, 201
38–39, 52, 142, 145, 153 blocks, 190
adaptive ranking, 104, 113–114, 116, 118–119, Bubble Rap, 159, 208–209, 212–214, 216–221
121–122, 124, 128, 130, 133, 135–136 Bubble Rap routing protocol, 177, 180
Adp, 119, 122, 129, 131–135, 131 buffer size, 90, 93
AdpOp, 122, 129–130, 131–135, 132 buffer time, 96, 98–99
AdpSpSynOp, 123, 129–131, 131–135, 133–135, bundle security protocols, 303
137
AdpSyn, 122, 131–135
aggregation, scheduling, and dropping, 16
algorithm and scoring function, 269 C
algorithms for transformation of tree structures, CBR, 43
260 centrality, 159
alleviate DTN Congestion, 32 central trusted authority, 188, 204
an opportunistic approach for mobile-­code-­based certification authority, 56–57, 59, 65, 75
distributed computing, 21–27 change degree of connectivity (CDC), 107, 120,
anonymous certificates, 57 122
anti-­localization techniques, 308 characteristics of opportunistic networks,
AODV, see ad hoc on-­demand distance vector 300–301
application binary interface (ABI), 192 city segment model, 231
application influenced movement model, 17–18 classification, 143
application program interface (API), 192 co-­location, 107, 120–121
asymmetric system, 56 community-­aware Bubble Rap, 105
attraction value, 110–111, 116 community-­based mobility models, 234
authentication, 56–60, 62, 64–68, 70–77 confidentiality, 56, 59–60, 69, 71, 77
availability, 56, 59, 61, 71 congestion, 301
average delay, 201, 204 connectivity, 301
average latency, 173, 183, 208–209, 215, 218, connectivity value, 110
220–221 considerations of small-­scale retransmission, 267
average remaining energy, 208–209, 215–217, contact duration (CT), 181–182
219, 221 context-­adaptive forwarding, 105, 138
context-­aware approach, 173, 176
B context-­aware routing (CAR), 177
context-­oblivious approach, 173, 174
bandwidth, 301 controlled replication, 210
Bellman-­ford, 41 control node, 194–195, 198–199, 203
betweenness centrality, 181 cooperative forwarder nodes (CFN), 188, 195,
binary spray and wait based on average delivery 198, 200
probability (ADPBSW), 175, 179 CORMAN fundamentals, 255

313
314 Index

cost, 104–105, 111–112, 114, 126–128, 130, energy efficiency, 209, 213, 215–216, 221
135, 137 EOC, 189, 193–195, 199
credit-­based reward, 188 Epidemic, 82–84, 104–105, 123, 128, 129, 130,
credit-­based schemes, 189 132–133, 131–135, 136, 157, 208–210,
cryptocurrencies, 188 213–214, 216–221
cryptography, 60, 71–72 epidemic routing, 174, 176–178, 180, 182–183,
185
erasure-­coding based routing, 175, 179, 185
D estimate/prediction routing, 178
damping factor, 106, 108–109, 124 Ethereum, 188–193, 196, 198–200
delay, 45–46, 48, 49, 51, 104, 112, 126–128, Ethereum virtual machine (EVM), 191
130, 135, 138, 142, 148, 149, 151, 153 evaluation of performance, 303
delay tolerant networks (DTN), 188–190, exchanged control messages, 127, 132–133, 133
192–197, 204–205 expected usage, 119–120, 122
delivery, 91–93 experiment settings, 274, 282, 287
delivery rate, 157, 167 extremely opportunistic routing, 245
delivery ratio, 104–105, 114, 126–131, 135, 137,
142, 148–150, 152, 181, 188, 201,
203–204, 208–209, 215, 217, 219–221 F
design of PSR, 256 first contact, 81–83
design of small-­scale retransmission, 268 F-­measure, 104–105, 126, 128, 129, 130, 135,
destination-­sequenced distance-­vector (DSDV), 136, 137, 139
38–39, 52 forwarder node, 188–190, 194–204
digital currency, 188, 190 forwarding, 81–82
Dijkstra algorithm, 84 fragment authentication, 305
dine and dash, 188, 195 fundamental algorithms–link state and distance
direct delivery, 81–83 vector, 253
direct transmission, 174–175, 179
disaster management, 189, 193–194, 204–205
dLife, 208–209, 212–214, 216–221 G
dLifeComm, 208–209, 212–214, 216–221
dropping and selective dropping attacks, 307 gas, 191, 193, 200, 202, 204
DSDV, 142–143, 153 gas consumption, 202, 204, 205
DTN, 81–82 general protocol architecture, 7–8
DTN lifetime control, 32–33 geographical restriction-­based mobility model,
dynamic multi-­routing, 15–16 230
dynamic prioritized scheduling, 33–34 grid job management, 24
dynamic routing and routing algorithm group signature, 57
deployment, 30–31
dynamic source routing (DSR), 38–39, 52
dynamic weight, 113, 128 H
harmonic mean, 126–128
E high error rate, 301
history-­based prediction routing (HBPR), 178,
EBubbleRap, 104–105, 123, 131–135, 132 180, 186
Edwards-­curve digital signature algorithm, 56, hop count, 96–98
63 hybrid erasure-­coding based routing (H-­EC), 175
effectiveness, 126–128, 129, 136, 137 hybrid routing protocol, 143
effectiveness performance index, 134, 134
effectiveness study of small-­scale retransmission,
281–286
effective path length (EPL), 177 I
efficiency, 126–128, 130, 131, 135–137 ICMN, 81
efficiency performance index, 127–128, 134, 135 implementation, 27–28
eight-­metric performance, 135, 136 implementation of reconstruction in differential
elliptic curve Diffie-­Hellman, 56–57, 62 update, 261
encounter-­and contact-­based routing, 159 incentive redemption, 195–196, 198, 204
energy consumption, 208–209 incentive schemes, 189, 197
 Index 315

integrating DTN WSNs in grid computer node speed, 178, 181

infrastructures using mobile code, non-­repudiation, 56–57, 59, 62, 65, 71, 77
22–24 NS-­2, 43, 84, 99, 142
integration value, 110, 111, 116
integrity, 56–57, 59, 61, 65–66, 71, 77
intelligent transport systems, 55 O
inter-­contact time (ICT), 181–182
interest awareness, 107–109, 113, 128, 137 observer node, 194–201, 204
interested forwarders, 126, 128, 129, 130, obstacle mobility model, 230
134–135 OMNet++, 86, 99, 142, 147
intermittent connectivity, 301 On-­Board Unit, 58
intermittently connected network, 190, 204 ONE, 82, 84–89, 99
inter-­probe time, 181 ONE simulator, 173, 182, 190, 199
IPeR, 108 OppComponent, 114, 115
IP multicast: a case example, 3–4 OppNet, 207–213, 215–216, 218–219, 221
issues in opportunistic networks, 301–302 opportunistic contact, 173, 182
opportunistic emergency scenarios applications,
28–33
K opportunistic network, 207, 299
Kalman filter, 107, 120 opportunistic network environment (ONE), 213
key management, 302, 304 opportunistic peer-­to-­peer network, 188, 197
kROp, 178, 180, 186 other variants of opportunistic forwarding, 250
overall performance of CORMAN, 286–290
overhead, 45–46, 48, 50, 51, 157, 167
L overhead ratio, 93, 95–96, 201, 204, 208–209,
large-­scale live update, 265 215–216, 218–221
large-­scale routing update and small-­scale
retransmission, 265–269
latency, 91, 94–95 P
link definitions on regular and sporadic links, packet delivery, 44, 47, 49, 51
165 packet loss, 142, 148,149, 151, 152
location index, 110–111, 116 parameters, 44
long delay, 301 pathway mobility model, 230
PeopleRank, 106, 108–109, 123, 131, 132,
M 131–135, 135, 138
performance index, 128, 134–135, 134–135
MANET, 141–142, 153 performance metrics, 209, 215, 221
Manhattan mobility model, 231 performance modeling for opportunistic
map-­based mobility models, 231 forwarding, 251
MaxProp, 84, 177, 185 Performance Study of PSR, 274–281
metrics, 43, 148 performance versus network dimension, 283, 288
mining, 191, 193 performance versus velocity, 286, 290
mobile ad hoc networks (MANETs), 37–38, 245 periodicity/pattern between node pairs, 160
mobile code, 4, 6, 13, 14 PIPeROp, 104–105, 108–109, 123, 129–131,
mobile opportunistic network (MON), 105, 107, 131–135, 134
112–113, 124, 138, 173 PISCAROp, 104–105, 108–109, 123, 129–131,
mobility, 300 131–135, 133, 135
mobility model, 42 PI-­SOFA Framework, 107–109, 130
motion vector (MoVe), 177, 180 place awareness, 113
multicast security, 303 point of interest, 84
popularity index, 111, 113
N post-­disaster communication network, 190, 193
power awareness, 107–109, 113, 117–119,
NAM, 43, 84–85 126–128, 130, 132, 136–138
neighborhood trimming, 257 power awareness performance index, 134–135,
network coding protocol, 175, 179 135
node density, 208–209, 214, 216–221 power consumption awareness, 131, 132
node remaining energy (NRE), 178, 181 power utilization fairness, 132, 132
316 Index

primitive services types and task delegation, scenarios suitable for opportunistic forwarding,
14–15 251
prioritized epidemic routing (PREP), 176 SCORP, 208–209, 212–214, 216–221
privacy, 56–59, 62, 68, 71, 75, 77 security, 302
private key, 56, 60, 62–66, 68–69, 71 architecture for opportunistic networks, 305
proactive protocol, 174, 179 attacks and their defense mechanisms,
proactive routing protocol, 142 306–309
proactive source routing (PSR), 256–261 considerations, 11–12
probabilistic protocols, 208, 211 initialization, 304
processing models, 25–26 issues in opportunistic networks, 302–303
processing time, 201, 204, 205 of routing protocols, 303
PRoPHET, 82–84, 158, 178, 180,183, 194–195, seek and focus routing protocol, 176, 179
200, 203, 208–209, 211–214, 216–221 select & spray, 111, 138
proposed routing with relay selection, 164 selfish behavior, 188, 194
protocols based on opportunistic data selfish forwarder nodes (SFN), 200–203
forwarding, 246 selfish and Sybil attacks, 308
public key, 56–57, 60, 62–66, 68, 70–71 shelter node, 194–195, 198–200
shortest expected path routing (SEPR), 177
shortest path, 84
Q shortest path map-­based movement model, 233
QoS, 37–38 similarity, 156
similarity interest, 107, 109, 115–117, 126
simulation, 147
R analysis, 234
random-­based mobility models, 228 results, 236
random direction model, 42, 84, 228 SLAW mobility model, 124, 138
randomized routing, 176, 179 small-­scale retransmission, 267
random walk model, 42, 84, 228 smart contracts, 191–192, 200, 204
random waypoint model, 42, 229 social-­based opportunistic forwarding, 104, 123,
RAPID, 84 136–137
reactive routing protocol, 143, 174, 179 social-­based protocols, 208–209, 212, 216
reification, 12–14 SocialCast, 106, 120
relay selection, 161 social graph metrics, 113
replay management, 302 social network-­based mobility models, 233
replication, 81 social network models, 234
replication-­based protocols, 209 social reference, 178, 181
revoked certificates, 56, 65, 76 solidity, 189, 192, 200, 205
rewarded/penalized similarity interest, 109 space syntax, 104–105, 110–113, 118, 125,
rewarded power, 108–109, 116–117, 121–122 129–131, 135–138
reward model, 189–190, 197–198, 203–204 metrics, 104, 110, 113, 116, 118, 123, 125, 125
road side unit, 56–59 opportunisitc forwarding, 110, 112
route update, 256 spatial dependency-­based mobility models, 230
route-­based map mobility model, 231 spray and focus, 84, 208–211, 213–214,
routing, 38, 301 216–221
routing algorithms, 173–174, 183 spray and focus routing protocol, 176, 180
The routing issue, 27 spray and wait, 84, 158, 174–175, 179–180,
routing and neighborhood update algorithm, 259 183,185, 208–211, 213–214, 216–221
RREQ, 39 stochastic mobility model, 228
RSCR: scheme design, 166 storage, 26–27
rush hour (human) traffic model, 231 storage constraint, 182
store-­carry-­forward, 81, 188–189, 204, 208
store-­process-­carry-­and-­forward paradigm,
S 24–25
streamlined differential update, 258
SAROS, 104, 124, 128, 136, 138
study on real-­life traces, 159
SCAR, 106–107, 109, 123, 129–131, 131–135,
suitability of existing routing protocols for
134–135, 138
opportunistic data forwarding in
scenario description, 4–6, 29–30
MANETs, 254
 Index 317

synthetic models, 229 U

system architecture elements, 8–11
UDP,
with density, 279
T with velocity, 281
uncontrolled replication, 209, 212
TCL, 84 uninterested users, 105, 114, 126, 129, 129
TCP usage profiles, 124
node density, 275 user activeness, 113
with velocity, 276 user density heatmap, 125, 125
temporal dependency-­based mobility models, user’s interest, 104, 106, 114–115, 117, 123,136
229 utility-­based routing, 176, 179
testing tool, 236
throughput, 45, 47, 48, 50, 52, 142, 148, 150,
152, 153 V
thrust areas of security, 303–305
timing strategy in routing protocols, 252 Vehicular Ad hoc Networks (VANETs), 38, 56
trace-­based model, 227–229
traffic investigation, 303 W
transaction, 188–191, 194–196
transient connectivity, 162 Weighted PageRank, 112
transmission latency, 157 Weights, 105, 107, 113–114, 116, 119, 136
Tree-­Based Routing Protocols Derived from the Working Day Movement Model, 232
Internet, 253 wormhole attacks, 306
trust management, 304
TTL, 39, 90, 93, 208–209, 214, 216, 219–221