doctor of philosophyshodhganga.inflibnet.ac.in/bitstream/10603/35832/1/final hemal sha… · hemal...
TRANSCRIPT
INVESTIGATIONS RELATING TO EFFICIENT DATA
TRANSMISSION AND RECEPTION IN AD-HOC
NETWORK
A
Thesis submitted in partial fulfillment of the requirements for
the Award of Degree of
DOCTOR OF PHILOSOPHY
To
Department of Computer Engineering
Faculty of Engineering & Technology
RESEARCH GUIDE:
Dr. Y. P. Kosta,Ph.D.
Director-Technical Campus
M.E.F.G.I., Rajkot
RESEARCH SCHOLAR:
Hemal Vinodkumar Shah
M.E. (C.E.)
Regn. No.: EN/008/009/2009
U.V. Patel College of EngineeringGanpat University, Ganpat Vidyanagar-384012
Gujarat, India
September 2013
CERTIFICATE
This is to certify that the thesis entitled “Investigations Relating to Efficient Data
Transmission and Reception in Ad-hoc Network” is submitted by Hemal
Vinodkumar Shah of U. V. Patel College of Engineering, Ganpat University is
bonafide work completed under my supervision and guidance for the Award of the
Degree of Doctor of Philosophy in the Faculty of Engineering & Technology,
Ganpat University, Gujarat, India.
The experimental work included in the thesis was carried out at Department of
Computer Engineering, U. V. Patel College of Engineering, Ganpat University and
the work is up to my satisfaction.
Research Guide:
Dr. Y. P. Kosta, Ph.D.
Director, Technical Campus,
Marwadi Education Foundation’s Group of Institutions
Rajkot-360 003. Gujarat, India
Forwarded through:
Dr. Paresh Shah, Ph.D.
Dean, Faculty of Engineering & Technology,
U.V. Patel College of Engineering
Ganpat University, Ganpat Vidyanagar-384012, Gujarat, India
Date: / /2013
Place: Ganpat University
CERTIFICATE
This is to certify that the thesis entitled “Investigations Relating to Efficient Data
Transmission and Reception in Ad-hoc Network” submitted by Hemal
Vinodkumar Shah fulfills the suggestions given by the Doctoral Committee during
pre-doctoral seminar held on 27th April, 2013 vide Ganpat University Letter No.
89/GNU/Ph.D./624/2013 dated 4th June, 2013 are duly incorporated in this thesis.
Dr. Y. P. Kosta, Ph.D.
Director, Technical Campus,
Marwadi Education Foundation’s Group of Institutions
Rajkot-360 003. Gujarat, India
Forwarded through:
Dr. Paresh Shah, Ph.D.
Dean, Faculty of Engineering & Technology,
U.V. Patel College of Engineering
Ganpat University, Ganpat Vidyanagar-384012, Gujarat, India.
Date: / /2013
Place: Ganpat University
THESIS APPROVAL SHEET
The Ph.D. thesis entitled “Investigations Relating to Efficient Data Transmission
and Reception in Ad-hoc Network” submitted by Hemal Vinodkumar Shah has
been approved for the Award of the Degree of Doctor of Philosophy under the Faculty
of Engineering & Technology, Ganpat University, Gujarat, India.
External Examiner: Research Guide:
( ) (Dr. Y. P. Kosta, Ph.D.)
Date:
Place: Ganpat University
DECLARATION BY THE CANDIDATE
I, Hemal Vinodkumar Shah, Regn. No. EN/008/009/2009, registered as a
Research Scholar for Ph.D. Programme in the Faculty of Engineering & Technology,
Ganpat University, do hereby submit my thesis, entitled “Investigations Relating to
Efficient Data Transmission and Reception in Ad-hoc Network” in printed as well
as electronic form for holding in the library of records of the University.
I hereby declare that:
1. The electronic version of my thesis in PDF format is submitted in CD-ROM.
2. The thesis is my original work and its copyright rests with me which does not
infringe or violate the rights of anyone.
3. The contents of the electronic version of my submitted thesis are the same as
those of the final hard copy submitted after my viva voce and adjudication of
my thesis.
4. I agree to abide by the terms and conditions of the Ganpat University Policy
with reference to Intellectual Property (hereinafter Policy) in force, as
approved by the competent authority of the University.
5. I agree to allow the University to make available the abstract of my thesis to
any user in both hard copies (printed) and electronic forms.
6. For the University’s own, non-commercial and academic use I grant to the
University the non-exclusive license to make limited copies of my thesis in
whole or part and to loan such copies at the discretion of the University to
academic persons and bodies approved from time to time by the University.
All usage under this clause will be governed by the relevant fair use provisions
in the Policy and by the Indian Copyright Act in force at the time of
submission of the thesis.
7. I agree to allow the University to place such copies of the electronic version of
my thesis on the private intranet maintained by the University for its own
academic community.
8. I agree to allow the University to publish such copies of the electronic version
of my thesis on a publicly accessible website on the internet.
9. If in the opinion of the University my thesis contains for patent or copyright
and if the University decides to proceed with the process of securing
copyrights and/or patents, I authorize the University to do so. I also undertake
not to disclose the patents or intellectual properties before being permitted by
the University to do so, or for a period of one year from the date of final thesis
examination, whichever is earlier.
10. In accordance with the Policy of the University, I accept that for any
commercial use of the intellectual property contained in my thesis, it is a joint
property of mine, my co-workers, my supervisors and the Institute. I authorize
the University to proceed with the protection of the intellectual property rights
in accordance with prevailing laws. I agree to abide by the provisions of the
University Rights Policy to facilitate the protection of intellectual property
contained in my thesis.
11. If I intend to file a patent based on my thesis when the University does not
wish so, I shall inform my intention to the University. In that case, my thesis
should be marked as patentable intellectual property and access to my thesis
should be restricted. No part of my thesis should be disclosed by the
University to any person(s) without my written authorization for one year after
my information to the University to protect the IP on my own, within 2 years
after the date of submission of the thesis or the period necessary for sealing the
patent, whichever is earlier.
Research Scholar:
Hemal Vinodkumar Shah
M.E. (Computer Engineering)
Regn. No: EN/008/009/2009
Date: / /2013
Place: Ganpat University
ACKNOWLEDGEMENT
In due process of obtaining the Doctorate Degree, submission of the thesis has been a
journey for which without the support of others it would not be possible.
Indeed, I am fortunate to have pursued my Ph.D. under the guidance of Dr.
Yogeshwar Prasad Kosta, the Director, Technical Campus, MEFGI, Rajkot. Dr. Kosta
is an eminent scholar in the areas of Telecommunication Engineering & Wireless
Technology, and at every step of my research he has been inspiring, motivating and
mentoring. Though I am failing in expression, but undoubtedly, I am indebted to him
in every sense. Let me declare that I have immensely benefitted by his profound
knowledge of the subject and in-depth understanding, and more than anything, else he
being a wonderful human being.
I am indebted to Shri Ganpatbhai Patel, the Patron-in-Chief, Shri Anilbhai Patel, the
President, Dr. L. N. Patel, the Vice Chancellor of Ganpat University; and with their
blessings I have been able to achieve my academic excellence.
I am extremely thankful to Dr. Naresh Jotwani, Deputy Director (R & D), Dr. Paresh
Shah, Dean, Faculty of Engineering & Technology, and to all the respected members
of the Doctoral Committee for their suggestions in shaping my research work.
For their kind and timely co-operation I am thankful to Shri J.V. Patel, Dr. Girish
Patel and Shri Jigar Raval of the University.
Let me express my sense of gratitude to the Head, Department of Computer
Engineering, UVPCE and my fellow-colleagues for their constant support to pursue
the research.
How could I forget Dr. D.C. Jinwala, Dr. N.J. Kothari, Dr. M. V. Joshi, Prof. Priyanka
Sharma, Dr. Kyunghan Lee, Dr. Pierre Ugo Tournoux, Dr. John Tang, Dr. Raghu
Raghavendra, Dr. Erik Kuiper,Dr. Satyen Parikh and Prof. Jignesh Mevada who have
helped me academically shaping my research work? They are such wonderful people
the more I write about them the less it is.
Without fail I must acknowledge Virkant, Ravi, Pravesh, Vishal, Jay, Hiloni and
Shalin, though they are my students but have been extremely helpful at every stage of
my research.
Chintan Upadhyay has remained a shadow-like with me throughout, without
mentioning him I would do a great injustice. He has extended unconditional support
during this period. There are friends and relatives, without naming I express my
sincere gratitude towards them.
My Dadi (Grandmaa), who is so close to my heart, though now suffering from old-age
ailments, with a touch of her hands I find solace and energy. I owe her so much.
I am indebted to my parents for their constant support and encouragement. Their
blessings have worked miracles in my life.
My greatest critic of my life, Dipal, my wife, has remained a rock-like throughout. It
is difficult to put in words about her sacrifice. Perhaps the Award of the Degree would
be more satisfying to her. My son - Kavish, even at his tender age could understand
the difficulties I have undergone during my studies, but has never disturbed or
demanded anything. Their unconditional love, affection and encouragement have
given me strength to take-up such a challenging work.
Hemal Vinodkumar Shah
DEDICATED TO MY FAMILY
i
INVESTIGATIONS RELATING TO EFFICIENT
DATA TRANSMISSION AND RECEPTION IN
AD-HOC NETWORK
ABSTRACT
In today’s world, connectivity via the Internet is an integral part of almost all the
devices and systems, whether mobile-device or otherwise, but the fact is, determined
connectivity is neither a rule nor mandatory practice. However, most wireless
applications demand stringent operating and connectivity attributes. For example,
most Micro and Pico networks Vehicular Networks, Pocket Switched Networks,
etc. operate in a hostile environment with weak-volatile links, node mobility, power
outages and interruptions that lead to frequent disconnections, but in episodic
networks, the network topology itself is the subject of change. Change attributes, such
as, time, rate, order and proportion, all these are difficult to predict, where, failure or
faults are not anomalies but rather an integral part of such opportunistic network.
Early studies on the network were on fixed network topologies, and in most analysis
the topologies remained invariant with time, i.e., topology-wise the network does not
change over time. Therefore, network parameters like calculating path length, cluster
coefficients and centrality were sufficient to determine the characteristic and
response. Currently, such approach to network analysis does not provide us the
complete behavioral information, primarily due to the dynamic nature of the network
which is constantly evolving, forming and dithering.
In this research, we successfully analyze and design the modeling aspects that help
predict the behavioural response of the challenged networks, characterized using our
model-approach. Attempts have been made to understand and evaluate temporal
properties of node’s mobility effectively. The study provides new insights which
helps in evaluating the behavior and characterization of network dynamics, thus
providing accurate information for routing. The information gained is quickly utilized
by injecting or spraying it back into the network to enable adaption. So as to ensure
that, the connectivity and operating conditions necessary for routing are maintained.
ii
That affects data exchange at appropriate levels in terms of speed and bandwidth. The
present study supports with evidence as to how this approach could minimize the
overhead and improve the delivery ratio?
We realize that in the world of information, mobility is fundamental. It plays a
significant role in information exchange in Intermittently Connected Mobile Ad-hoc
Networks where traditional end-to-end path do not exist. Therefore, we recognize the
fact that in the present wireless environment, nodes are mobile, and to our advantage,
a possible solution can be explored. One way of doing this is to look for opportunities
and timely strike to create a channel for communication and information exchange. In
our research, by exploiting mobility of the nodes connectivity can be established and
ad-hoc-channels are operated.
The proposed model of Intermittently Connected Mobile Ad-Hoc Network is
presented using time varying graphs, characterized temporal distance and temporal
centrality. This model captures delay, contact duration, frequency and time order of
contacts compared to metrics constructed for analyzing static network conditions. By
using these intertwine parameters; matrix is constructed that provides an insight to
help evaluating the relationship between critical network parameters and design a
temporal characterization algorithm. Our designed algorithm is applied for the
purpose of validation to RollerNet, INFOCOM real traces and Random Way Point
synthetic dataset as proof of concept, calculating network parameters such as; number
of time frame, time window size and temporal distance, temporal closeness centrality
etc. The parameter values are utilized by an Adaptive Routing protocol for
understanding the dynamics of networks and taking meaningful decisions. The study
explores achieving effective performance in terms of delivery probability, the number
of dropped messages, overhead ratio, latency of delivered message etc. under the
assumption that a network is secured and is free from any transmission errors. The
findings of our study reveal that there is a significant improvement in delivery ratio
after the application of the concept.
Keywords: Intermittently Connected Mobile Ad-hoc Networks, Time Varying
Graph, Adaptive Routing, Routing Efficiency, Forwarding, Replication, Node
Mobility
iii
INVESTIGATIONS RELATING TO EFFICIENT
DATA TRANSMISSION AND RECEPTION IN
AD-HOC NETWORK
ABSTRACT
Connectivity via the Internet has become an integral part of today’s life. It is to
connect, share and communicate information through all the available devices and
systems, regardless of being a part of infrastructure or infrastructure-less network. As
known that determined connectivity is neither a rule nor a practice because most
wireless applications demand stringent operating attributes. The wireless environment
is comprised of Micro and Pico networks like Vehicular and Pocket-Switched
networks. These are dynamic in nature and operate in hostile environments where
parameters constantly change in terms of time, rate, order and proportion resulting in
losses. Because of the known-unknown, or unknown-unknown reasons this makes
network behavioural-response difficult to predict; as failure or faults are not
anomalies but an integral part of the dynamic network. In this research, we analyze
and design the modeling aspects that predict the behavioural-response of Intermittent
Networks. This research exploits mobility associated with the nodes that establish
connectivity and affect data transfer. The model, Intermittently Connected-Mobile
Ad-Hoc Network is realized by using time varying graphs, temporal distance and
temporal centrality. Accordingly, a temporal algorithm is designed to characterize the
response of the network. This algorithm is utilized to validate real traces and synthetic
datasets by calculating the number of time-frames, time window size and temporal
distance, temporal closeness centrality etc. The parameters obtained are utilized to
improve the routing efficiency. The Adaptive Routing is designed which utilizes the
temporal properties for accurate forwarding decisions. It integrates Encounter Based
Forwarding and Two Period Spray & Wait based’ Replication. The research
establishes that there is a significant increase in delivery ratio and reduction in
overhead improving routing efficiency.
Keywords: Intermittently Connected-Mobile Ad-hoc Network, Time Varying Graph,
Routing Efficiency, Forwarding, Replication
iv
TABLE OF CONTENTS
ABSTRACT ................................................................................................................. i
TABLE OF CONTENTS .......................................................................................... iv
LIST OF TABLES ................................................................................................... viii
LIST OF FIGURES ................................................................................................... ix
LIST OF ABBREVIATIONS ................................................................................... xi
NOTATIONS............................................................................................................. xii
Chapter 1. Introduction ............................................................................................1
1.1 Routing Challenges in IC-MANET ....................................................3
1.2 Need for Research...............................................................................5
1.3 Defining the Problem..........................................................................5
1.4 Objective and Scope of Research .......................................................6
1.5 Contributions.......................................................................................7
1.6 Chapter Outline...................................................................................8
Chapter 2. Related Work ..........................................................................................9
2.1 Characteristics of IC-MANET............................................................9
2.2 Routing in IC-MANET.....................................................................11
2.2.1 Classification Based on Knowledge Available at Nodes......12
2.2.2 Classification Based on Number of Message Carriers .........12
2.2.3 Popular IC-MANET Routing Algorithms ............................13
2.3 Real Networks Changes Over Time .................................................20
2.3.1 Online Social Networks ........................................................21
2.3.2 Technological Networks .......................................................23
Chapter 3. Analysis of Density Aware Routing And Graph Theory ..................27
3.1 Analysis of Density Aware Spray & Wait ........................................28
3.1.1 Delivery Ratio Comparison of DASW and Epidemic usingRollerNet...............................................................................29
3.1.2 Delivery Ratio Comparison of DASW and Epidemic usingINFOCOM’05.......................................................................31
3.1.3 Delivery Ratio Comparison of DASW and Epidemic usingINFOCOM’06.......................................................................32
3.2 Static Graph and Metrics ..................................................................33
3.2.1 Graph Metrics .......................................................................34
3.2.2 Centrality...............................................................................36
v
Chapter 4. Temporal Algorithm.............................................................................39
4.1 Temporal Graphs ..............................................................................39
4.1.1 Temporal Metrics..................................................................41
4.1.2 Time Window Calculation....................................................43
4.1.3 Preconditions.........................................................................44
4.2 Temporal Algorithm .........................................................................46
4.2.1 Temporal Betweenness Centrality ........................................47
4.2.2 Temporal Closeness Centrality .............................................48
4.3 Application of Temporal Algorithm .................................................50
4.3.1 Experimental Dataset and Common Format.........................51
4.3.2 Time Window Size Calculation for Experimental Datasets .53
4.3.3 Temporal Closeness Centrality Evaluation...........................55
4.4 Observations .....................................................................................56
Chapter 5. Adaptive Routing..................................................................................58
5.1 Analyzing Modified DASW .............................................................59
5.1.1 Performance Evaluation of Modified DASW usingINFOCOM’05.......................................................................59
5.2 Forwarding........................................................................................60
5.2.1 Algorithm: Encounter based Forwarding..............................61
5.3 Replication ........................................................................................61
5.3.1 Algorithm: Two Period Spray & Wait..................................62
5.4 Algorithm: Adaptive Routing ...........................................................62
5.4.1 Average Temporal Distance Gain.........................................63
5.5 Temporal Closeness Centrality based Approach ..............................64
Chapter 6. Performance Evaluation ....................................................................66
6.1 Evaluation Approach based on Virtual Test Bench ..........................67
6.1.1 Assumptions..........................................................................67
6.1.2 Datasets .................................................................................67
6.1.3 Performance Metrics .............................................................68
6.1.4 Input Parameter Settings in ONE Simulator.........................68
6.2 Graphical Analysis and Representation of the Findings...................70
6.2.1 Delivery Ratio Comparison of AR with RollerNet vsRWP_63................................................................................71
6.2.2 Overhead Ratio Comparison of AR with RollerNet vsRWP_63................................................................................71
6.2.3 Number of Dropped Messages Comparison of AR withRollerNet vs RWP_63...........................................................72
vi
6.2.4 Average Latency of Delivered Message Comparison of ARwith RollerNet vs RWP_63 ..................................................72
6.2.5 Delivery Ratio Comparison of AR with INFOCOM’06 vsRWP_98................................................................................73
6.2.6 Overhead Ratio Comparison of AR with INFOCOM’06 vsRWP_98................................................................................74
6.2.7 Average Latency of Delivered Messages Comparison of ARfor INFOCOM’06 vs RWP_98 .............................................75
6.2.8 Delivery Ratio Comparison of AR with Spray & Wait andPRoPHeT using INFOCOM’06 and RWP_98 .....................75
6.2.9 Overhead Ratio Comparison of AR with Spray & Wait andPRoPHeT using INFOCOM’06 and RWP_98 .....................76
6.2.10 Comparing Number of Dropped Messages in AR with Spay& Wait and PRoPHeT using INFOCOM’06 and RWP_98..77
6.2.11 Comparing Average Latency of Delivered Messages in ARwith Spray & Wait and PRoPHeT using INFOCOM’06 andRWP_98................................................................................78
6.2.12 Comparing Delivery Ratio in AR with Spray & Wait andPRoPHeT using RollerNet and RWP_63 .............................79
6.2.13 Comparing Overhead Ratio in AR with Spray & Wait andPRoPHeT using RollerNet and RWP_63 .............................80
6.2.14 Comparing Number of Dropped Messages in AR with Spray& Wait and PRoPHeT using RollerNet and RWP_63..........80
6.2.15 Comparing Average Latency of Delivered Message in ARwith Spray & Wait and PRoPHeT using RollerNet andRWP_63................................................................................81
6.2.16 Comparing Delivery Ratio in AR with Spray & Wait andPRoPHeT under RollerNet and RWP_63 using TemporalCloness Centrality.................................................................82
6.2.17 Comparing Overhead Ratio in AR with Spray & Wait andPRoPHeT under RollerNet and RWP_63 using TemporalCloness Centrality.................................................................83
6.2.18 Comparing Number of Dropped Messages in AR with Spray& Wait and PRoPHeT under RollerNet and RWP_63 usingTemporal Closeness Centrality .............................................84
6.2.19 Comparing Average Latency of Delivered Messages in ARwith Spray & Wait and PRoPHeT under RollerNet andRWP_63 using Temporal Closeness Centrality....................85
Chapter 7. Summarization and Conclusion .........................................................87
7.1 Summarization ..................................................................................87
7.2 Conclusion ........................................................................................88
7.3 Proposed Future Work ......................................................................89
References ...............................................................................................................91
vii
Appendix – A..............................................................................................................98
Appendix – B ............................................................................................................101
Appendix – C............................................................................................................103
Appendix – D............................................................................................................117
List of Publications ..................................................................................................121
viii
LIST OF TABLES
Table 2.1 Comparisons of Traditional Network and IC-MANET.........................10
Table 2.2 Summary and Comparison of IC-MANET Routing Schemes...............18
Table 2.3 Summary of Empirical Datasets with Temporal Information................25
Table 3.1 Abacus: Node Degree vs Message Copies.............................................29
Table 3.2 Node Degree Measurement of Synthetic and Real Trace......................29
Table 3.3 Internal Contact and Average Node Degree ..........................................30
Table 3.4 Computation of Shortest Path ................................................................35
Table 3.5 Computation of Clustering Coefficient..................................................35
Table 3.6 Computation of Degree Centrality.........................................................36
Table 3.7 Computation of Closeness Centrality ....................................................37
Table 3.8 Computation of Betweenness Centrality ...............................................38
Table 4.1 Interaction Sequence Between Nodes....................................................40
Table 4.2 Time Window Calculation.....................................................................43
Table 4.3 Temporal Centrality Calculations ..........................................................50
Table 4.4 Experimental Datasets ...........................................................................52
Table 4.5 Common Format for Experimental Datasets .........................................52
Table 4.6 Time Window Calculations for RollerNet, INFOCOM’06 and RWP...53
Table 4.7 Temporal Distance for RollerNet, INFOCOM’06 and RWP ................54
Table 4.8 Closeness Centrality for RollerNet, INFOCOM’06 and RWP ..............55
Table 6.1 ONE Settings for INFOCOM’06 and RollerNet ...................................69
Table 6.2 ONE settings for RollerNet and RWP_63 .............................................69
Table 6.3 ONE settings for INFOCOM’06 and RWP_98 .....................................70
Table B-1 Summary of Real and Synthetic Dataset Collections ..........................101
Table D-1 Average Temporal Distance Values for RWP_63 ...............................117
ix
LIST OF FIGURES
Figure 1.1 Snapshots of IC-MANET at Four Different Times .................................4
Figure 2.1 IC-MANET Routing Algorithms Classification....................................11
Figure 2.2 Exchange of Summary Vectors for Epidemic Routing..........................13
Figure 3.1 Average Delay for SnW(N) vs Average Node Degree ..........................28
Figure 3.2 Message Interval vs Delivery Ratio of DASW and Epidemic Routingusing RollerNet......................................................................................30
Figure 3.3 Message Interval vs Delivery Ratio of DASW and Epidemic Routingusing INFOCOM’05..............................................................................31
Figure 3.4 Message Interval vs Delivery Ratio of DASW and Epidemic Routingusing INFOCOM’06..............................................................................32
Figure 3.5 Static Graph ...........................................................................................33
Figure 4.1 Temporal Graph ....................................................................................40
Figure 4.2 Betweenness Centrality..........................................................................47
Figure 4.3 Betweenness Centrality by Consideration of Time Duration ................48
Figure 4.4 Temporal Closeness Centrality ..............................................................49
Figure 4.5 Temporal Metrics Evaluation and Utilization Scenario.........................51
Figure 5.1 Message Event Interval vs Delivery Ratio for Modified DASW usingINFOCOM’05 .......................................................................................59
Figure 6.1 Delivery Ratio Comparison of AR with RollerNet vs RWP_63 ...........71
Figure 6.2 Overhead Ratio Comparison of AR with RollerNet vs RWP_63..........71
Figure 6.3 Message Drop Comparison of AR with RollerNet vs RWP_63............72
Figure 6.4 Average Delivered Message Latency Comparison of AR withRollerNet vs RWP_63 ...........................................................................73
Figure 6.5 Delivery Ratio Comparison of AR with INFOCOM’06 vs RWP_98 ...73
Figure 6.6 Overhead Ratio Comparison of AR with INFOCOM’06 vs RWP_98..74
Figure 6.7 Average Delivered Message Latency Comparison of AR withINFOCOM’06 vs RWP_98 ...................................................................75
Figure 6.8 Delivery Ratio Comparison of AR with Spray & Wait and PRoPHeTusing INFOCOM’06 and RWP_98 .......................................................76
Figure 6.9 Overhead Ratio Comparison of AR with Spray & Wait and PRoPHeTusing INFOCOM’06 and RWP_98 .......................................................77
x
Figure 6.10 Comparing Number of Dropped Messages in AR with Spay & Wait andPRoPHeT using INFOCOM’06 and RWP_98 ......................................77
Figure 6.11 Comparing Average Latency of Delivered Messages in AR with Spray& Wait and PRoPHeT using INFOCOM’06 and RWP_98 ..................78
Figure 6.12 Comparing Delivery Ratio in AR with Spray & Wait and PRoPHeTusing RollerNet and RWP_63 ..............................................................79
Figure 6.13 Comparing Overhead Ratio in AR with Spray & Wait and PRoPHeTusing RollerNet and RWP_63 ...............................................................80
Figure 6.14 Comparing Number of Dropped Messages in AR with Spray & Waitand PRoPHeT using RollerNet and RWP_63 .......................................81
Figure 6.15 Comparing Average Latency of Delivered Message in AR with Spray& Wait and PRoPHeT using RollerNet and RWP_63 ..........................82
Figure 6.16 Comparing Delivery Ratio in AR with Spray & Wait and PRoPHeTunder RollerNet and RWP_63 using Temporal Closeness Centrality .83
Figure 6.17 Comparing Overhead Ratio in AR with Spray & Wait and PRoPHeTunder RollerNet and RWP_63 using Temporal Closeness Centrality ..84
Figure 6.18 Comparing Number of Dropped Messages in AR with Spray & Waitand PRoPHeT under RollerNet and RWP_63 using Temporal ClosenessCentrality ...............................................................................................84
Figure 6.19 Comparing Average Latency of Delivered Message in AR with Spray& Wait and PRoPHeT under RollerNet and RWP_63using TemporalCloseness Centrality ..............................................................................85
Figure C.1 ONE Connectivity Report....................................................................105
xi
LIST OF ABBREVIATIONS
ACC Average Closeness Centrality
AR Adaptive Routing
AVG Average Closeness Centrality per Time Window
CRAWDAD Community Resource for Archiving Wireless Data At Dartmouth
CWC Current Window Counter
DASW Density Aware Spray & Wait
DTN Delay Tolerant Network
DTNRG Delay Tolerant Network Research Group
EV Encounter Value
MANET Mobile Ad-hoc Networks
Mbps Mega Bits Per Second
MB Mega Bytes
MF Message Ferries
NS2 Network Simulator 2
IC-MANET Intermittently Connected -Mobile Ad-hoc Networks
IP Internet Protocol
OSN Online Social Networks
ONE Opportunistic Network Simulator
PRoPHeT Probabilistic ROuting Protocol using History of Encounters and
Transitivity
PSN Pocket Switched Networks
RW Random Walk
RWP Random Way Point
TCP Transmission Control Protocol
TTL Time To Live
SnW Spray & Wait
Wi-Fi Wireless Fidelity
WLAN Wireless Local Area Network
WWW World Wide Web
xii
NOTATIONS
C Clustering Coefficient
c2 Cost as Number of Copies used Per messageC Degree Centrality of Node iC Closeness Centrality of Node iC Betweeness Centrality of Node id (T , T ) Temporal Distance between Nodes i and j between Tmin and Tmax
E Set of Edgesg Temporal Graph g with Time Window w
k Number of time units elapsed
L Number of Message Copies
L* Average Path Length
L1 Number of Message Copies to Spray in the First Time Period
L2 Number of Message Copies to Spray in the Second Time Period
Ni Neighbour of Node i
P(a,b) Probability at every node ‘a’ for each destination ‘b’
Pinit Initialization Constantp Temporal Path starting at Node i and ending at j with h Hopes
¬SVB Message note in Summary Vector of Node B
Tmin Start Time
Tmax End Time
td Time Deadline
Ux(Y) Utility of node x for destination Y
V Set of Nodes
(Vt,Et) Set of Nodes Vt and Edges Et in Time Window t
w Time Window Size
β Scaling Constant
γ Ageing Constant
Chapter 1
IntroductionCHAPTER 1. INTRODUCTION
1. Introduction
1INTRODUCTION
The wireless network has become increasingly popular in the communication
industry; communication per se has intruded into our daily life, and always unsettled
our dormant mindset which is constantly striving magical things to happen. Like or
dislike its influence has become unavoidable. With the changing times these so called
network provides mobile users with ubiquitous computing1 capability and information
access regardless of the users’ location.
Currently we use two variations of mobile wireless networks: (i) Infrastructure (ii)
Infrastructure-less networks. In the former, it has a fixed and wired gateway or fixed
Base-Stations connected through wires, and in the latter the mobile nodes are
connected within the range of base-station, where “Hand-off” travels out of range
from one station to another, and the host is able to continue communication.
If we talk of the infrastructure-less network which is often called as ‘Mobile Ad-hoc
Network’ (MANET), it has no fixed routers, where every node is a router, and are
capable of movements dynamically connected network in an arbitrary manner. In this,
1 http://www.itu.int/osg/spu/ni/futuremobile/pervasive/index.html
1. Introduction
the responsibility of organizing and controlling the network lies within the
distribution of nodes. Thus, the entire network remains mobile and individual nodes
move freely, though occasionally some pairs don’t communicate directly with each
other but rely upon the other nodes for message delivery. This is also known as multi-
hop networks, mostly found in air and space, or on very small devices in which the
nodes can function as routers.
The use of MANET is mostly envisioned in closed (sensor) networks and other
controlled networks i.e., tactical deployments, though not yet widespread
elsewhere[1]. Two technical inhibitors for MANET are identified as below:
As we are aware that running Internet Protocols(IP) between mobile peers require an
end-to-end path to be found and sustained for a sufficient period of time, to allow the
respective application interactions to complete. In MANETs, the existence of such a
path certainly depends on density of mobile nodes and movement dynamics.
Likewise, using the most dominant Transport Control Protocol (TCP) over a series of
wireless (ad-hoc) links (e.g., in wireless meshes or MANET) results in poor
performance [2] with an increasing number of hops.
This suggests that both aspects are being addressed by the ad-hoc network approach
of Intermittently Connected Mobile Ad-hoc Networks (IC-MANET), also known as
Delay Tolerant Networks (DTN) [3]. The concept of IC-MANET is already in use in
the areas like Battlefield, Internet Services to Remote Areas[4][5],Wildlife
Monitoring, Whales Tracking[6],Oceans, Lake Water Quality Monitoring2, Online
Social Networks(OSN) and Human Contact Networks. This is to deal with frequent
disconnection, high error rate, long and variable delays and asymmetric data rates
because in IC-MANET, the existence of end-to-end connection cannot be assumed.
2 http://down.dsg.cs.tcd.ie/sendt/
1. Introduction
1.1 Routing Challenges in IC-MANET
There are limitations in IC-MANET, despite of the fact that the connectivity of nodes
is not constantly maintained, but still, it is desirable to allow communication between
nodes to happen. Therefore, it becomes a necessity to provide a routing protocol
which tries to route packets opportunistically when link is available. But this cannot
be done by traditional protocols as it assumes that the entire network is connected all
the time.
In a traditional network the routing protocol computes the path to the destination
assuming that there is a high reliability of delivery. However, routing in IC-MANET
is challenging as the nodes are mobile and connectivity is episodic.
The transient network connectivity is a primary concern in the design of efficient
routing algorithms for IC-MANET. Therefore, the routing of the packets is based on
Store-Carry-Forward[7] paradigm. That is, when a node receives a message if there is
no path to the destination or connection to any other node, the message is buffered.
And the upcoming opportunities to meet other nodes are waited. Even if a node meets
with other node, it carefully decides whether to forward the message or buffer it.
Replicating a message to multiple nodes increases the delivery probability. However,
this may not be the right choice as it causes message overhead and additional resource
consumption. Whereas, keeping a single copy of message and forwarding it to a few
numbers of nodes it uses the network resource efficiently. But, in it the message
delivery probability becomes lower and delivery delay gets higher. Thus, there is a
tradeoff among the message delivery ratio, overhead ratio and delivery delay. Hence,
while designing an efficient routing protocol for IC-MANET, the important
considerations are:
1. The number of copies[8] distributed to the network for each message, and
2. The selection of routing technique replication[9] or forwarding[10].
1. Introduction
Given the collections of measurements related to large real network traces, the
researchers are realizing that connections are inherently varying over time and exhibit
more dimensionality than static analysis can capture.
Let’s consider the time varying intermittent network, illustrated in Figure 1.1, which
presents four different snapshots of the network showing connectivity between nodes
at four different times. Assume node ‘A’ has a message destined to node ‘G’.
Looking at the snapshots, it is observed that delivery of the message could be
achieved by node ‘B’ at T4 if node ‘A’ forwards the message to node ‘B’ at time T1.
Here the key point is how node ‘A’ will know when node ‘B’ will meet the
destination? This makes routing challenging in IC-MANET.
Let’s study the routing issues in intermittent network and analyze how time varying
graph[11] can help in designing an efficient algorithm.
Figure 1.1 Snapshots of IC-MANET at Four Different Times
1. Introduction
1.2 Need for Research
In the backdrop of technological advancements where there is a massive empirical
and evolving data, the analysis of real, social, biological and technological networks
has become challenging. In such scenario including IC-MANET understanding the
dynamics of the relationship, contact duration, repeated occurrence of contact etc
have remained under explored3 with regard to time varying properties [12] By
extending the same concept in IC-MANET, the efficiency of routing protocol can be
improved .This can help understanding the dynamic behavior and making appropriate
routing decisions. Therefore, the present research provokes exploring different
possibilities by utilizing the same time varying analysis. During the investigation
some observations have been made:
1. Are there any temporal properties [13],[14],[15] analyzed or proposed for IC-
MANET?
2. Can the time varying behavior of mobile ad-hoc network be used for improving
the routing efficiencies?
3. Is it possible to integrate Forwarding and Replication[8]?
The above questions have tempted for further investigation, defining the different
metrics related to temporal graph leading to Proposed Routing Protocol.
1.3 Defining the Problem
An attempt is made to analyze, understand and evaluate temporal networks for
defining the time varying properties that enables to characterize network dynamics
and provide accurate information for data diffusion. It explores how quickly it adapts
3 Had a discussion with Dr. K. H. Lee, one of the pioneers of IC-MANET. He was of the view thatthere is an ample scope of research for utilizing the temporal properties in t design of IC-MANETrouting. Refer appendix-A excerpts of communication.
1. Introduction
the network conditions, and makes the routing decision accordingly by minimizing
the delay, overhead and improving the delivery ratio.
1.4 Objective and Scope of Research
The objective of this research work is to design and develop an efficient routing
protocol for IC-MANET by using the temporal properties[16]. We intend to achieve
an efficient performance in terms of delivery probability, number of transmissions,
overhead ratio and average latency of delivered message within available resources
under an assumption that the network is secured and is free from any transmission
errors.
The scope of research is as under:
1. Understanding temporal distance and temporal centrality that captures delay,
duration and time order of contacts (interactions).
2. Design and Development of the Temporal Characterization Algorithm.
3. Applying, Evaluating and Extending the Temporal Algorithm to real traces and
synthetic datasets.
4. Design, Analysis and Implementation of Adaptive Routing(AR) using
Opportunistic Network Environment(ONE) simulator[17].
5. Performance Analysis and Comparision of AR with Spray & Wait[9] and
Probabilistic Routing Protocol using History of Encounters and Transitivity
(PRoPHeT)[18].
1. Introduction
1.5 Contributions
We have introduced the temporal graph model, formalized the notion of shortest
paths, distance and centrality used in graph theory for IC-MANET. It has shown that
the static graph underestimates the time order of contacts, frequency, repeated
occurrence and the shortest paths. Contrary to the intuition, it is seen that the evolving
graph [11] can be instrumental in evaluating above-mentioned parameters.
By designing the temporal algorithm we have analyzed and evaluated the temporal
properties of real trace4 (RollerNet, INFOCOM’05, INFOCOM’06) and synthetic5
datasets (Random Way Point(RWP)). These properties are used in designing and
developing routing protocols for improving the routing efficiencies.
A Temporal Algorithm is proposed to characterize temporal properties which
ascertains:
1. Temporal Algorithm captures temporal distance and temporal centrality
2. The designed and developed Temporal Algorithm is capable of evaluating any
experimental dataset given in a prescribed format.
3. Using the Temporal Algorithm real traces like RollerNet, INFOCOM’06 and
RWP synthetic datasets are evaluated for temporal properties.These properties
are used for designing efficient Adaptive Routing.
4. AR is designed and developed by using temporal properties, in this:
a. AR integrates Encounter Based Forwarding and Two Period Spray &
Wait.
b. Routing engine dynamically calls Forwarding and Replication schemes
based on average temporal distance gain per time window.
5 For the following scenarios performance analysis is carried out using ONE
simulator:
4 Real trace is those mobility patterns that are observed in real time networks.5 Synthetic dataset attempts to realistically represent the behaviour of mobile nodes without the use oftraces.
1. Introduction
a. AR is measured for delivery ratio, overhead ratio and latency of
delivered message while comparing with RollerNet vs RWP and
INFOCOM’06 vs RWP.
b. AR is compared with the Spray & Wait and PRoPHeT for delivery
ratio, overhead ratio, number of dropped messages and latencies.
1.6 Chapter Outline
Outline of the thesis is as under:
It is seen that IC-MANET changes over time, therefore existing static network
analysis cannot fully capture dynamics. Chapter 2 takes into account of IC-MANET,
its Characteristics, Applications, Challenges and Routing. Chapter 3 is an overview
of Static and Dynamic Graphs: Algorithms, theories, methodology, and the
conceptual system design that underpin this work. Chapter 4 discusses about
modeling of IC-MANET as Time Varying Graphs, Application of Temporal Metrics
and its related Time Window Computation, Conditions and Design philosophy of
Temporal Characterization Algorithm with examples, and Application of Temporal
Algorithm to evaluate experimental dataset. Chapter 5 reveals Analysis and Design
of Adaptive Routing for Encounter based Forwarding and Two Period Spray & Wait.
Chapter 6 presents scenarios using the simulation environment and performance
analysis of AR. It also discusses the measures taken in terms of delivery ratio,
overhead ratio, latency and the number of dropped messages. Chapter 7 draws
conclusion from the foregone study, evaluating and discussing the findings,
contributions, and limitations of this investigation highlighting for further research in
the same area.
Chapter 2
Related WorkCHAPTER 2. RELATED WORK
2. Related Work
9
2RELATED WORK
Today’s Internet is highly successful and has achieved worldwide adoption. Existing
network architectures, communication models and network protocols are sufficient
and efficient in use under normal circumstances. However, there are regions and
situations where no fixed network infrastructure is available. These ad-hoc networks
are frequently partitioned and there is no guarantee of continuous and stable
connectivity. The network devices in such scenarios are often constrained by their
transmission range, processing power, storage space and energy. In such
environments, conventional networking protocols perform poorly[19] and are often
not suitable.
2.1 Characteristics of IC-MANET
As discussed above , a number of research initiatives have been undertaken in recent
years, primarily under the umbrella of the Delay Tolerant Network Research Group6
6 http:// www.dtnrg.org/wiki
2. Related Work
10
(DTNRG) and a general architecture class called the Delay Tolerant Network(DTN)
[3]. It covers all the aspects of mobile ad-hoc networks. In the MANET connectivity
is episodic, periodic or opportunistic, therefore, DTN is opportunistic or challenged,
extreme, partially connected or partitioned, or Intermittently Connected Mobile Ad-
hoc Network (IC-MANET). Since the IC-MANET makes limited assumptions about
end-to-end connectivity and peer resources, it utilizes node mobility, message caching
and relaying techniques to achieve asynchronous and message-oriented
communication.
The characteristics of IC-MANET are :
It has a high level of heterogeneous nodes;
Intermittent connectivity;
Endures recurrent interruption and failures;
Has asymmetric, long and variable data rates;
Suffers from energy, bandwidth, storage/memory and cost constraints.
In traditional network, end-to-end paths exist between source and destination to
transfer data, whereas in IC-MANET, such a path cannot be assumed. The mobile
nodes, when come into each other’s transmission range, they connect, and as they go
away from that range resulting in connectivity disruption, called Intermittent
Connectivity. Here, node acts as a router having persistent storage to support Store-
Carry-Forward operation over multiple paths and over large timescales. TCP/IP works
on some fundamental assumptions built into the Internet’s architecture makes it
unsuitable[19] for IC-MANET.Comparison is summarized as presented in Table 2.1
below:
Table 2.1 Comparisons of Traditional Network and IC-MANET
Characteristic Traditional Network IC-MANET
End-To-End Connectivity Continuous Intermittent Connectivity
Propagation Delay Short Long
Transmission Reliability High Low
Link Data Rate Symmetric Asymmetric
2. Related Work
11
Because of the abovementioned characteristics of IC-MANET, there are challenges
like routing, resource allocation, security, architecture and addressing; out of these
routing is important.
2.2 Routing in IC-MANET
The main objective of routing is to maximize the probability of message delivery and
minimize the resource consumption (i.e., Buffer Space, Network Bandwidth and
Battery Energy) and overhead. Although the IC-MANET applications are expected to
be tolerant of delay, does not mean that they would not benefit from decreased delay,
but it is still meaningful to minimize the delivery latency.
It has been almost a decade since Kevin Fall [20] talked about routing issues in
partially connected networks. Since then many studies have been taken place as to
how to handle the sporadic connectivity between nodes. However, different
classifications of these algorithms can be made with reference to number of carriers
and information available at the node. In Figure 2.1 two different classifications
derived from [21] is shown.
Figure 2.1 IC-MANET Routing Algorithms Classification
2. Related Work
12
2.2.1 Classification Based on Knowledge Available at Nodes
In some studies[20], it is assumed that each node in the network has exact knowledge
of (past and future) node trajectories, or node meeting times and durations. Therefore,
the messages are routed over pre-determined paths. But these algorithms, which
assume the existence of oracles giving further information, are unrealistic, because the
intermittent connectivity between the mobile nodes does not allow nodes to have such
information. There are also a number of significant studies (such as Epidemic[22],
PRoPHeT[18], Spray & Wait[9][22]) assuming zero knowledge about the
aforementioned feature of the nodes. These algorithms either forward the messages
randomly or use the meeting history of nodes (which can be obtained locally through
encounters with other nodes) and forward over different paths in a non-deterministic
manner.
2.2.2 Classification Based on Number of Message Carriers
There are classifications based on number of message carriers. In algorithms such as
SCAR[23], MaxProp[24] and Delegation Forwarding [25] there exist only one node
carrying the message at all times. In this, the messages are forwarded to nodes having
higher chance to meet the destination. There are methods commonly used for multiple
carriers, in some, a number of copies generated and distributed to multiple nodes to
increase the delivery probability. In others [8][9][22], a limited copies are distributed.
P. Mundur et al. [26] supports that there is flooding like dissemination of the message
copies.
There are schemes [27][28] use Erasure Coding technique for efficient routing. It first
processes and converts a message of k data blocks into a large set of Ф blocks. The
original message is reconstructed from subset of Ф blocks, and then encoded blocks
are distributed to other nodes expecting delivery of sufficient number of blocks.
Let’s discuss the popular routing algorithms.
2. Related Work
13
2.2.3 Popular IC-MANET Routing Algorithms
The pioneering algorithm in the field of multi-copy is Epidemic[29]. This protocol
behaves epidemic-like algorithms [30][31]. In this, when the nodes get into contact,
an exchange of a pair-wise information happens resulting delivery of message, and a
summary vector is transferred to the other node as it holds the index of the message.
Then, each node learns the message IDs which are not available in its own buffer and
requests to transfer from the other node. However, in the absence of path the message
gets buffered. Figure 2.2 illustrates the same.
Let’s understand the procedures from host A’s point of view. The same procedure
applies to host ‘B’. First ‘A’ transmits its summary vector SVA (a compact
representation of all the messages being buffered at ‘A’) to ‘B’. Next, ‘B’ performs a
logical AND operation between the negation of its summary vector¬SVB (represents
the messages that it needs) and SVA. By this way, host ‘B’ determines the difference
between the messages buffered at ‘A’ and ‘B’, then, it transmits a vector requesting
these messages from ‘A’. Finally, host ‘A’ transmits these requested messages to host
‘B’. This message exchange procedure is always applied when two hosts come into
contact; as a result, if the buffer space at hosts and contact time is sufficient the
messages are delivered. Thus, the study reveals that Epidemic Routing has got the
highest delivery ratio, but suffers from wastage of network resources due to
Replication.
Figure 2.2 Exchange of Summary Vectors for Epidemic Routing
To reduce such wastage of network resources Lindgren et al. have proposed
PRoPHeT [18]. They believed that in MANET the nodes move in a predictable
fashion based on repeating movement patterns. If a node has visited a location several
times before, it is likely that it will visit that location again. Depending on this the
2. Related Work
14
delivery rate of messages is aimed to improve while keeping the buffer usage and
communication overhead at a low level.
PRoPHeT algorithm operates in a similar way as Epidemic does, means when two
nodes meet, they exchange summary vectors as Epidemic. But an additional piece of
data, called delivery predictability is also exchanged between the nodes. It is defined
as P(a,b)∈ [0, 1], at every node ‘a’ for each known destination ‘b’, indicating how the
node ‘a’ will be able to deliver a message to ‘b’. When this exchange is done,
according to summary vectors, the messages are requested from the other node. The
difference between PRoPHeT and Epidemic Routing is about forwarding the strategy:
the former allows transferring the message based on delivery predictability values of
encountered node and the latter it simply replicates.
To calculate the value of delivery predictability, there are three steps:
Inspired by non-random mobility of nodes which also provides different meeting
popularities to different nodes, the authors preferred the most encountered nodes
among the others for the delivery of messages. Therefore, to reflect this property
in the delivery probability metric of a node, P(a,b) is updated whenever a node is
encountered, so that nodes that are encountered to have a higher delivery
predictability. This calculation is shown below, where Pinit ∈ [0,1] is an
initialization constant.P( , ) = P( , ) + (1 -P( , ) ) x P (2.1)Where, P(a,b)ϵ[0,1]
If a pair of node does not encounter for a while, they are less likely to be a good
forwarders. Therefore, the authors added an aging mechanism as γ ∈ [0, 1)
representing γ as the aging constant, and k is the number of time units that have
elapsed since the last time the metric was aged. It shows the time unit used in the
formulation can differ, and should be defined based on the application and the
expected delays in the targeted network.P( , ) = P( , ) x γ (2.2) The last one is to measure the transitivity in which node ‘a’ frequently meets node
‘b’ and node ‘b’ frequently encounters node ‘c’, suggesting that the node ‘a’ is a
good candidate to relay message to c (through b) even if it rarely meets ‘c’.
2. Related Work
15
P( , ) = P( , ) + (1 -P( , ) ) x P( , ) x P( , ) x β (2.3)Where, β is the scaling constant to decide the impact of transitivity on the delivery
predictability. Therefore, the forwarding happens only when the delivery
predictability of neighboring node is higher.
Like PRoPHeT, Harras et al. [22] have used probabilistic modeling for message
forwarding. They have added additional information on top of the probabilistic model.
It shows the impact of controlled message flooding schemes over sparse mobile
networks on message delay and network resource consumptions.
Li et al [32] have proposed an efficient Store-Carry-Forward based scheme called
Adaptive Multi-Copy Routing for packet delivery in IC-MANET. They have stated
that instead of using a source-defined replication factor each relay node decides to
replicate a message. In other words, the number of message copies distributed in the
network is not decided at source in the beginning, but decided by each individual
node according to the current network conditions and end-to-end delay. Thus the
approach becomes a replication-factor-free and more cost-efficient than the copying
schemes.
In a routing protocol presented by Jones et al. [33], a single copy of the message is
generated and forwarded using the predicted topology information. Burgess et al. [24]
have proposed MaxProp, a routing protocol based on prioritizing both the schedule of
packets transmitted to other nodes, and the schedule of packets is deleted from the
buffer. In MaxProp, the construction of a real life DTN test-bed is discussed where
the transfer duration or available buffer space is limited. It ranks the packets
according to the given criteria to handle overflow and transfer duration.
An algorithm proposed by T.Spyropoulos et al. [9] performs fewer transmissions than
flooding based routing schemes and delivers message faster than the existing single
and multi-copy schemes. For this purpose, there are two different routing algorithms
discussed under (1) Spray &Wait (SnW) (2) Spray & Focus:
2. Related Work
16
1. Spray & Wait: The authors try to combine the efficiency of flooding based
algorithms and the simplicity of Direct Transmission. There are two phases:
Spray Phase: In this, a limited number of message copies (L) are spread over
the network from source. If the destination is one of the relay nodes then the
message is delivered, otherwise, relay nodes enter into wait phase. Based on
the number of message copies to be sprayed various schemes have been
surveyed[34].
Wait Phase: Each node carrying a message copy tries to deliver to the
destination via Direct Transmission[35].
2. Spray & Focus: In above scheme source sprays message copy blindly to each
encounter node and performs Direct Transmission in the wait phase. To overcome
this, a different version of Spray & Wait algorithm has been proposed [36]:
Spray Phase: In this, every message originating at the source node, (L)
message copies are spread to L distinct nodes. It is the same as given in the
spray phase of above algorithm.
Focus Phase: Once the spraying phase is over, the nodes start roaming to find
the destination. Each message copy is forwarded via node’s utility index[10].
If UX(Y) denotes the utility of node ‘X’, for destination ‘Y’, then a node ‘A’
having a copy of the message destined for node ‘D’, forwards its copy to a
new node ‘B’ in its range, if and only if UB(D) > UA(D) + Uth, here, Uth
denotes the utility threshold parameter.
On the other hand, to make routing more reliable, there are routing algorithms based
on Erasure Coding [37]. In this, Wang et al. have presented the advantages
(robustness to failures etc.) over the Replication. The split of erasure coded blocks
over multiple delivery paths (contact nodes) to optimize the probability of successful
message delivery is referred[38]. A similar approach focusing on the distribution of
encoded blocks among the nodes is also presented by Liao et al[27]. They have
2. Related Work
17
proposed Estimation based Erasure Coding on assumption that the nodes in the
network are not identical.
As an extension of this work Zhang et al [39], also utilizes the information on a
node’s available resource (buffer space, remaining energy level etc.) in the evaluation
of the node’s capability to successfully deliver the message. Hybrid Routing is
proposed combining the strengths of Replication and Erasure Coding[40]. In addition
to encoding each message into a large amount of small blocks, these blocks are
replicated to increase the delivery rate.
Let’s discuss an opportunistic network, where nodes are human carrying devices,
mostly found in social networks. Here, the contacts (i.e. message exchange
opportunities) between these mobile devices depend on the social relationships.
Researchers have utilized the community detection, similarity, betweenness etc. to
understand the contact patterns between nodes to develop better routing algorithms.
As in social network there are communities formed by nodes which meet more
frequently than the nodes outside their communities, such formation affects the
routing decisions.
Daly et al. [41] have used the betweenness and similarity metric to increase the
routing performance. When two nodes contact, they calculate the utility function
comprising of these two metrics. Then, the node having higher utility value is given
the message. In BubbleRap[42], each node is assumed to have two rankings: global
and local, the former denotes the popularity (i.e. connectivity) of the node in the entire
society, the latter denotes own community. The messages are forwarded to nodes
having higher global ranking until a node in the destination’s community is found.
Then, the messages are forwarded accordingly. By this way, first the probability of
finding the destination’s community is increased. After the message reaches the
destination’s community, the probability of meeting the destination increases, so that
the shortest delivery delay is attempted. In Publish/Subscribe [43], many-to-many
communication paradigm is addressed as an extension to end-to-end style, usually
assumed in IC-MANET. Then, using the centrality values of nodes, an effective
multi-point communication and efficient routing is enabled.
2. Related Work
18
In Community-based Epidemic Forwarding scheme, called LocalCom[44], the
community structure of the network is detected by using local information of nodes.
The message is forwarded to each community through gateways. Also there are
studies where several interesting properties of social networks are considered like
irregular deviations from the habitual activities of the nodes[45]. It is shown that the
worst-case performance of routing can be improved by scattering multiple copies of a
message, in a way that deviant (less frequently encountered) nodes are closed to at
least one of these copies. The effects of socially selfish behaviour of nodes on routing
have been studied [46].
Summary of Routing Schemes
The summary and comparison of routing schemes[21] are shown in Table 2.2. The
Mobility Model for simulation denotes the movement patterns of mobile nodes
designed to approximate the real situations for evaluating the protocol. The item of
Applicable Environment expresses the limitations of the protocol as quoted by the
authors. In this section, we have tried to summarize the popular routing schemes for
IC-MANET. For extensive study readers interested in this field may refer the surveys
[8][47].
Table 2.2 Summary and Comparison of IC-MANET Routing Schemes
Sr.No.
Protocol No. ofMessageCopies
Main Strategy Real Trace /SyntheticDataset used?
ApplicableEnvironment
1. Direct Delivery[2][35]
Single Source waits until itcomes into contact
Not Available Info stationArchitecture
2. First Contact[20]
Single Use any availablecontact. No oracleused.
Remote village,city bus networkscenario
General DTN
3. Message Ferries[48]
Single MF takes charge offorwarding
MF follows therectangle route,and other nodesadopt randomwaypoint orarea-basedmodel
Ferries move in aproactive mannerforcommunication
4. Throwboxes[49]
Single Deploy static relaydevices based ondifferent levels ofinformation toenhance datatransfer capacity
RWP andUMass model
Static nodes areneeded forenhancingnetworkconnectivity
2. Related Work
19
5. Seek and Focus[10]
Single Make use ofrandomizedforwarding andutility based routing
RWP andCommunitybased mobility
General mobilenetwork
6. MobySpace[50] Single Find the node withthe similar patternof mobility as thatof destination forforwarding themessage
Power-law basedmovementpattern
Assume thatmobility patternof destination isknown.
7. Epidemic [29] Multiple Flood message RWP General mobilenetwork
8. PRoPHeT [18] Multiple Select the relaynodes by Predictingthe deliveryprobability
RWP andCommunitybased model
General mobilenetwork
9. Spray &Wait[9]
Fixed Take advantage ofthe limited numberof message copiesto replicate
RWP andCommunitybased model
Prefers thenetwork withsufficient mobilenodes
10. Spray & Focus[9]
Fixed Take advantage ofthe limited numberof message copiesto replicate
RWP andCommunitybased model
Prefers thenetwork withlocalized nodes
11. Island Hopping[51]
Multiple Rely on the clusterto forward message
Random Walk(RW)
Rely on thepresence of stabletopology ofclusters
12. SimpleReplication[33]
Multiple The source of themessage generatesmultiple copies,while the relaynodes are onlyallowed to send tothe destination
ZebraNet Wildlifemonitoring
13. History BasedReplication[33] [38]
Multiple Source creates “r”identical copies of amessage, which arethen delivered to thebest “r” nodes,where quality isdetermined byhistory.
ZebraNet Wildlifemonitoring
14. ErasureCoding[37]
Multiple k*r fragmentstotaling “r” timesthe message size aregenerated and sentto the first k*rintermediate relays.
ZebraNet Generic routingscheme
15. EstimationBased ErasureCoding [27]
Multiple Two intermediatenodes exchangedata until thenumber offragments for thedestination isproportional to thenodes’ probabilityof meeting thedestination
RWP Generic routingscheme
2. Related Work
20
From the above Table, it is observed that many existing routing protocols use
opportunistic approach to pass messages via every possible contact at a future time,
and utilizes synthetic datasets for various performance measurements. But, underlying
time information7 is rarely explored for routing decisions. The following section
discusses in detail about temporal information hidden in real network traces, and such
analysis can reveal accurate information for efficient data dissemination.
2.3 Real Networks Changes Over Time
There are ranges of empirical networks studied more recently possessing temporal
information. The taxonomy presented is by no means exhaustive but demonstrates the
types of inherent temporal information available in the datasets. The available and
collectable temporal network data are summarized in Table 2.3.
7 Time information related to timestamps, duration, frequency and time-order.
16. MV Routing[52]
Multiple Find peer having amaximumprobability ofvisiting the regionof destination
Synthetic datasetof peermovements in ageographic area.
Vehicular areanetwork
17. MaxProp[24] Multiple Forward themessage to anydevice having amaximumprobability ofdelivering themessage to thedestination
Map basedmobility
Vehicular areanetwork
18. EarliestDelivery[20]
Multiple Compute the pathusing modifiedDijikstra algorithm
Not available FutureKnowledge
19. EncounterBased Routing[53]
Fixed Compute node’sencounter valuewith neighbor nodeand decide based onthat forward numberof message copies
RW and RealTraces
Encounter valueper node can beutilized in thesingle copyscheme to selectnext relay node
20 Multi PeriodSpray &Wait[54]
Fixed First spray copiessmaller thannecessary. Ifdelivery does nothappen sprayadditional copies.
RW,RWP andRollerNettraces
Investigate formore realisticmobility modelsand for highlyvarying nodebehavior
2. Related Work
21
2.3.1 Online Social Networks
The popularity of websites that allow us to keep in touch with friends has exploded
over the last decade. The convenience of maintaining friendship networks online,
sending messages to friends, arranging events, uploading photos and sharing locations
and thoughts have produced household brands such as Facebook and Twitter.
Since these services are online, the massive data needs to be stored and maintained by
the service providers. This data becomes a valuable asset for marketers and
researchers. Typically, researchers are not able to access the data from the service
providers, but may access by writing a programme for academic purposes. This
practice is known as crawling has enabled helping key studies for analysis of large
Online Social Networks(OSN) datasets:
Facebook: Being a very popular OSN there exists several different crawled datasets8
[55][56] However, the most comprehensive is that of the University of Santa
Barbara[57] which includes both the social network and interactions between users. In
this, an individual user profile provides information on their friends, and similarly
their information is shared among themselves. This suggests, we can construct the
social network of people (nodes) and their relationships (edges). In interaction
networks, individuals can post messages on each other’s profile pages; this again
creates an interaction network (nodes represent people and a directed edge represents
a message being sent).
Temporal Information: As users add new friends and delete the old ones, the social
network topology changes over time. In the case of interactions, messages are
timestamped so information about interactions at different times is available.
Twitter: Twitter is a service where users can share a short message consisting of 140
characters, known as ‘tweets’. Users subscribe to (or ‘follow’) other users profile to
8 The Facebook Project.” [Online]. Available: http://www.thefacebookproject.com
2. Related Work
22
access their tweets but, unlike other OSN such as Facebook, friendship do not need to
reciprocate. Recent datasets have crawled the entire corpus of tweets over a months’
period [58] and a subset of tweets that included tweet location information over a
twelve-day period [59]. The former dataset contained 41.7 million user profiles, 1.47
billion directed links and 106 million tweets; whereas the latter datasets which filtered
out users with geographic information contained 400,000 user profiles, 183 million
directed links and 334.5 million tweets. Both these datasets allow us to construct two
different graphs: a graph of followers and a graph of tweets.
Temporal Information: Both datasets contain timestamps of each tweet and can be
traced the dynamic spread of tweets as it cascades through the user network. The
users can constantly follow or do not, the topology of followers’ changes over time.
2.3.2 Human Contact Networks
The study of close-range human contacts have received attention from technologists
interested in opportunistic routing in Pocket Switched Networks(PSN)[42]. This has
resulted in several experiments aiming to record participants' meetings with each
other. In the Haggle [60][61], participants experimented by carrying Bluetooth
enabled devices that scanned and recorded other Bluetooth devices in proximity
(within a 30 meter range). In this, different environment and number of participants
were used, ranging from an office with 12 users to a conference with 78 users, with
the intention of investigating decentralized routing.
In the Reality Mining [62], 100 participants were given a Bluetooth enabled
smartphone to carry on campus over a period of 9 months for data mining human
social behaviour such as predictability. The devices recorded other Bluetooth devices
in proximity.
In Rollerblading Tour[63] 2,500 users participated, the duration of the tour was about
3 hours and deployed contact loggers on 62 volunteers. In addition to the loggers,
other participants were asked to activate their Bluetooth. They used iMotes and
Bluetooth to periodically log (every 15 seconds) the encounters and established that
2. Related Work
23
every time a logger discovered a neighboring device, it stored the time and identity of
the latter.
All the above studies allow us to infer when a pair or group of people are in proximity
(for radio communication or to transmit a biological virus), the graph of people
(nodes) and their contacts (edges) can be generated.
Temporal Information: Since the timestamps are logged, the information on
duration of a meeting, periodic patterns and inter-contact time can be analyzed over a
period of time.
2.3.2 Technological Networks
The technological networks generally store data in digital form that can be retrieved.
World Wide Web (WWW): The web contains billions of web pages, can hyperlink
to (and be hyperlinked from) other pages. This produces a network of web pages
(nodes) and directed hyperlinks (edges) to study scale-free networks [64]. Such data is
regularly crawled by search engines such as Google, Yahoo! Bing etc. by following
and recording hyperlinks from one web page to another. So that search results are
informed by the popularity of certain web pages available for research.
Temporal Information: The WWW changes frequently as web pages and hyperlinks
are added or deleted; therefore, there is rich information on the changing topology of
the web.
Wi-Fi Hotspot: These are ubiquitous in homes, offices, campuses, airports and high
streets. With the devices regularly connecting to Wi-Fi routers that log the access.
Such logs are used to identify device locations and infer co-locations to generate a
graph of devices (nodes) and co-locations (edges). Numerous studies have been
performed to collect data on different time scales, spatial properties and environments
in campus over 5 years from 450 Wi-Fi access points[65], office environment over a
2. Related Work
24
week across 151 access points[66], and city environment over 3 years from numerous
free access points Montreal[67].
Temporal Information: Timestamps are associated with a device connecting to (and
disconnecting from) an access point. This gives us connection duration, co-location
with other devices and possible periodic movement pattern (e.g., a user connecting at
particular time and location every day in the office).
Digital Communication Networks: We can communicate globally via emails
through a centralized server. The server provides us with data on interactions between
the users; from this data; a network of users (nodes) and the messages sent (edges) are
constructed. There are studies analyzing scale-free properties of 57,000 University
email users [68], corporate emails between 151 colleagues during a 3 year period
before a corporate filing bankruptcy [69] and six-degrees of separation between users
on a global scale instant messenger service with 30 billion conversations between 240
million users [70].
Temporal Information: Messages between users are timestamped, and instant
messenger sessions (i.e. the time two users are engaged in conversation) are engaged
over some time
Through these real network datasets, we can isolate four distinct sources of time
information:
1. Timestamps: It can be associated with both nodes and edges.
2. Durations: Are implicit in these timestamps, e.g., how long two people meet?
3. Frequency: It uncovers patterns in edge or node occurrences, and periodicity is
present in certain datasets like daily meetings with colleagues etc.
4. Time-Oder: Time dependency between events.
2. Related Work
25
Table 2.3 Summary of Empirical Datasets with Temporal Information
Temporal InformationAvailable?
Datasets Nodes Description Datasets
Available?
Source
Facebook,Friendship
6m Addition anddeletion of friends
No -
Facebook,Interaction
6m Time of interaction No -
Twitter 41.6m Time of tweet Yes -
INFOCOM’05* 78 Time of contact, startand end time
Yes CRAWDAD9
INFOCOM’06* 98 Time of contact, startand end time,topological changes
Yes CRAWDAD
RollerNet** 62 Time of contact, startand end time, Stringeffect, topologicalchanges
Yes CRAWDAD
Reality Mining 100 Time of contact, startand end time
Yes CRAWDAD
WWW 153127 Topological changesover time
No -
Chapter Summary
In this chapter, different routing schemes for IC-MANET mobile network have been
presented and popular schemes are discussed with details. The study reveals that in
opportunistic schemes there are concerns for number of message copies and adaption
between Forwarding and Replication. Delivery ratio and other performance related
metrics under synthetic and real datasets have been evaluated for routing. But, the
study gives a clear indication that the time information relating to real network data
analysis is underestimated. The following chapter unfolds in a greater length
9 CRAWDAD is the Community Resource for Archiving Wireless Data At Dartmouth, a wirelessnetwork data resource for the research community* Indicates datasets use for experiments
2. Related Work
26
interpreting the static graph theories for identifying the important metrics, and how it
redefines for time varying graphs?
The existing literature has helped exploring the temporal properties for improving the
routing efficiencies. The study has further intimidated to support examining the time
varying information for understanding timestamps, frequency, duration and time order
Table 2.2 in the given chapter indicates how most (though it is not exhaustive) of the
schemes are evaluated for delivery ratio and other performance related metrics?
Chapter 3
Analysis of DensityAware Routing And
Graph TheoryCHAPTER 3. ANALYSIS OF DENSITY AWARE
ROUTING AND GRAPH THEORY
3. Analysis of Density Aware Routing And Graph Theory
27
3ANALYSIS OF DENSITY AWARE ROUTING AND
GRAPH THEORY
In the previous chapter, broad classification of routing into a single copy and multi
copy has been presented. It is seen that there is a trade-off between the message
delivery ratio, overhead ratio and delivery delay in the networks amongst Forwarding
and Replication.
Important considerations for designing an efficient routing protocol are: the number
of message copies, selection of appropriate forwarding, and or replication strategy.
Limited message copies represent a controlled replication based scheme[8], and
choosing the right number of copies is an open issue[9], whereas, selection of
appropriate forwarding or replication depends on routing objectives like delivery
ratio, delay, overhead , latency.etc.
Our analysis and investigation is focused on the technique of density aware controlled
replication and static graph. The study also discusses on the need of temporal
information for improving routing efficiency.
3. Analysis of Density Aware Routing And Graph Theory
28
3.1 Analysis of Density Aware Spray & Wait
Under real traces, challenges and observations the Density Aware Spray & Wait
(DASW) are being analyzed. Pierre et.al.[63] analyzed the dynamics of a real trace
collected in a pipelined10 DTN during Rollerblader tour11. They have experimented
accordion phenomenon12 through the variation of average node degree having a major
impact on the performance of epidemic disseminations. Based on expected average
delivery delay requirement and node degree, the right amounts of message copies are
chosen.
As shown below in Figure 3.1, each node using DASW[63] performs a looks up in
abacus. This abacus returns number of message copies as per the requirement based
on target average delay and node degree. For example, to achieve an average delay of
150 seconds with an average node degree 6, abacus suggests spraying four message
copies.
Figure 3.1 Average Delay for SnW(N) vs Average Node Degree
10 Pipelined-DTN is particular class of IC-MANET characterized by a one-dimensional topology.11 Discussed in Section 2.3.2- Human Contact Network12 Accordion phenomenon means topology that expands and shrinks.Also, it is known as “stringinstability or slinky effect”.
3. Analysis of Density Aware Routing And Graph Theory
29
In selecting DASW scheme for analysis, there are two reasons:
1. It presents a recent technique for selection of message copies based on network
density i.e.average node degree.
2. It uses RollerNet traces for analyzing ‘slinky effect’: Topology that expands and
shrinks with time.
3.1.1 Delivery Ratio Comparison of DASW and Epidemic usingRollerNet
DASW uses spraying number of copies in the network as shown in Table 3.1. For
RollerNet traces the average node degree varies from 5 to 12, and the number of
copies to spray varies from 128 to16 according to abacus.
Table 3.1 Abacus: Node Degree vs Message Copies
Degree Number ofcopies
5.0 128
5.2 100
5.4 80
5.8 64
Degree Number ofcopies
6.0 50
6.6 40
7.0 32
7.5 29
Degree Number ofcopies
8.0 25
9.2 20
10.0 18
11 16
The above table helps us measuring the average node degree for Shortest Map Based
Movement model and RWP .In Table 3.2, node degree varies from 0 to 6 for Shortest
Path Map Based Movement, suggesting, spraying needs 50 to 128 copies
Table 3.2 Node Degree Measurement of Synthetic and Real Trace
Dataset Node DegreeRollerNet 6 to 12ShortestPathMapBasedMovement 0 to 6RWP 0 to 3
3. Analysis of Density Aware Routing And Graph Theory
30
We have measured the delivery ratio of DASW and Epidemic using RollerNet traces.
It is observed in Figure 3.2 that DASW requires around 128 message copies as node
degree varies from 5 to 12. But, there are only 62 nodes present in trace, resulting in
spraying more number of message copies than the number of nodes, behaving as
Epidemic Routing.
Figure 3.2 Message Interval vs Delivery Ratio of DASW and Epidemic Routingusing RollerNet
It is evident that DASW performs better with RollerNet. If we change the real trace it
behaves like Epidemic Routing. Let’s see in the following experiments:
RollerNet, INFOCOM’05 and INFOCOM’06 have been selected for simulation. Here
total number of internal contacts, duration and average node degree for the three are
summarized in Table 3.3.
Table 3.3 Internal Contact and Average Node Degree
INFOCOM’05 INFOCOM’06 RollerNetDuration (days) 3 4 0.12Internal Contacts 38530 23310 49946Average Node Degree 0.051 0.208 5.5
00.10.20.30.40.50.60.70.80.9
1
5 10 15 20 25
Del
iver
y R
atio
Message Intervals
Message Intervals vs. Delivery Ratio
DASW
Epidemic
3. Analysis of Density Aware Routing And Graph Theory
31
It is learned that the average node degree is below 6 for Rollernet, suggesting that the
number of copies required to spray are around 128. Further, the numbers of nodes
participating in INFOCOM’06 are 98, and in Rollernet it is 63. Thus, indicating that
the number of message copies spread is more than the number of nodes, resulting
Epidemic like behavior.
3.1.2 Delivery Ratio Comparison of DASW and Epidemic usingINFOCOM’05
Figure 3.3 Message Interval vs Delivery Ratio of DASW and Epidemic Routingusing INFOCOM’05
As shown in Figure 3.3 DASW performs like Epidemic because of very low
neighborhood degree. Eventually, the number of copies spread is more than the
number of nodes.
0
0.01
0.02
0.03
0.04
0.05
0.06
5 10 15 20 25
Del
iver
y R
atio
Message Intervals
Message Intervals vs. Delivery Ratio
DASW
Epidemic
3. Analysis of Density Aware Routing And Graph Theory
32
3.1.3 Delivery Ratio Comparison of DASW and Epidemic usingINFOCOM’06
Figure 3.4 Message Interval vs Delivery Ratio of DASW and Epidemic Routingusing INFOCOM’06
DASW performs poorly with INFOCOM’06 datasets due to lower node degree and
fewer internal contacts as shown in Figure 3.4 suggesting that to achieve expected
delay, higher numbers of copies are required, resulting in wasting the network
resources behaving like Epidemic.
Observation and Need for understanding Static Graph Theory:
The study reveals that the average node degree calculation is completely static and
based on abacus. Also DASW performs well with RollerNet.
1. It is understood that the RollerNet traces contains inherent temporal information,
however, the analysis of time information is completely ignored in DASW.
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
5 10 15 20 25
Del
iver
y R
atio
Message Intervals
Message Intervals vs. Delivery Ratio
DASW
Epidemic
3. Analysis of Density Aware Routing And Graph Theory
33
2. The collected traces were only the current snapshot of the complete timeline of
the network, Where temporal information is available, it is assumed that authors
explicitly have ignored time information, and have constructed a static graph
from the union of edges across all temporal occurrences for expected delay and
delivery computations.
The above analysis raises some fundamental questions as to which metrics to follow
from static graph theory, and how does this theory and analysis be used for re-
defining temporal graph? The following section addresses defining the metrics
relating to static graph, namely path, shortest path length and centralities.
3.2 Static Graph and Metrics
The network that do not change over time or built as a result of aggregation of
information over a certain period of time, is called static network, representing static
graph. But it is observed [13] that connections in IC-MANET are inherently time
varying and exhibit more dimensionality[16] than static analysis can capture.
Figure 3.5 Static Graph
Let’s consider the static graph as shown in Figure 3.5, the shortest path from A to G is
through A,B,D,E,G, and the shortest path length (A,G) = 4 hops, assuming that
3. Analysis of Density Aware Routing And Graph Theory
34
1. the links are always available,
2. nodes are available.
And to understand the static graph as defined:
Graph G is 2-tuple (V, E) where, V is the set of nodes (or vertices) and E is the set of
edges (or links) connecting a pair of nodes, therefore, a graph can be directed or
undirected where an edge between two nodes is either non-mutual or symmetric. Also
a graph can be weighted or unweighted.
Directed or undirected graph is represented as N-by-N adjacency matrix A, where N =|V|, and the value aij at row i and column j is non-zero if an edge exists from node i to
j.
In the case of an unweighted graph aij = 1 if there is an edge, 0 otherwise; in the case
of a weighted graph aij can be any real number. The following section defines metrics
for undirected and unweighted graph.
3.2.1 Graph Metrics
Two exponents, Watts & Strogatz [71] have formalized the path length and the
clustering coefficient metrics.
Paths and Shortest Path Length
Before defining the characteristics of path length, it is imperative to analyze the
concepts of paths and path lengths. Let’s take Pij in a path denominated as a list of
nodes starting from node i and finishing at j, where an edge exists between each
intermediate pair of nodes and the length of a path is measured by the number of
intermediate hops from source to destination. All paths are acyclic i.e., there are no
cycles or repeated nodes.The shortest (or geodesic) path length, dij from ‘i’ to ‘j’ is
defined as the minimum path length L over all paths. From this, the characteristic (or
average) path length, L* is defined as:
3. Analysis of Density Aware Routing And Graph Theory
35
L*= ( )∑ d (3.1)Where, N is number of nodes, V is set of nodes (i,j).
Table 3.4 Computation of Shortest Path
A B C D E F G H Average Path Length L*
A 0 1 1 2 3 4 4 4 0.34B 1 0 1 1 2 3 3 3 0.25C 1 1 0 2 3 4 4 4 0.34D 2 1 2 0 1 2 2 2 0.21E 3 2 3 1 0 1 1 1 0.21F 4 3 4 2 1 0 1 2 0.30G 4 3 4 2 1 1 0 1 0.29H 4 3 4 2 1 2 1 0 0.30
Clustering Coefficient
It measures the number of nodes that are direct neighbours. More formally, for a node
‘i’, its clustering coefficient Ci is calculated as the fraction of links that exists between
the neighbours of a node ki of node ‘i’, over the total possible number of edges ki (ki -
1) /2. From this, the average clustering coefficient of a graph, C, is defined as:
C = 1N C (3.2)For the Static Graph presented in Figure 3.5 the following table shows the
computation of clustering coefficient for each node.
Table 3.5 Computation of Clustering Coefficient
A B C D E F G H Clustering Coefficient CA 0 1 1 2 3 4 4 4 0.34B 1 0 1 1 2 3 3 3 0.25C 1 1 0 2 3 4 4 4 0.34D 2 1 2 0 1 2 2 2 0.21E 3 2 3 1 0 1 1 1 0.21F 4 3 4 2 1 0 1 2 0.30G 4 3 4 2 1 1 0 1 0.29H 4 3 4 2 1 2 1 0 0.30
3. Analysis of Density Aware Routing And Graph Theory
36
3.2.2 Centrality
In complex network and social network analysis, centrality refers to the identification
of the most important nodes. Node importance is interpreted in many ways depending
on their applications, e.g., persons with many friends can deliver messages more
quickly.The centrality is defined depending upon three categories such as: degree,
closeness and betweenness[72].
Degree Centrality
It is chosen based on the popularity of nodes measuring number of neighbours of node
‘i’ of Ni = ∑ aij.
For the degree centrality of a ‘i’ is defined as the number of neighbours Ni of i,
normalized by the maximum number of distinct connections:
C = NN − 1 (3.3)For the Static Graph presented in Figure 3.5 the following table shows the
computation of degree centrality for each node.
Table 3.6 Computation of Degree Centrality
Degree Centrality Cideg
A 0.29B 0.43C 0.29D 0.29E 0.57F 0.29G 0.43H 0.29
3. Analysis of Density Aware Routing And Graph Theory
37
Closeness Centrality
It measures how quickly a node communicates with all other nodes. This is calculated
for a node ‘i’ as the average shortest path length, d, to all other nodes. This is defined
as C = NN − 1 d (3.4)For the Static Graph presented in Figure 3.5 the following table shows the
computation of closeness centrality for each node.
Table 3.7 Computation of Closeness Centrality
Closeness CentralityCi
clo
A 5.43B 6.00C 5.43D 3.43E 6.86F 4.86G 6.86H 4.86
Betweenness Centrality
It measures[72] the shortest paths that passes through a node, considered to be the
proportional flow of data passing through each node. The betweenness of node ‘i’ is
calculated as the proportional number of shortest paths between all node pairs that
passes through i. This is being defined as:C = P (i)/P, (3.5)Where,
Pi,j is the number of shortest paths starting from source node ‘i’ and destination node
‘j’ and Pj,k (i) are those paths which pass through node ‘i’
A key point is that the betweenness takes into account of alternative shortest paths
which is crucial in measuring the robustness. If node ‘i’ is the only bridging node on a
3. Analysis of Density Aware Routing And Graph Theory
38
path then its removal would be highly detrimental, whereas, if there are another path
not including ‘i’ then its role becomes less critical.
For the Static Graph presented in Figure 3.5 the following table shows the
computation of betweeness centrality for each node
Table 3.8 Computation of Betweenness Centrality
Betweenness CentralityCi
bet
A 0.25B 0.61C 0.25D 0.68E 0.7F 0.25G 0.2H 0.25
Chapter Summary
The chapter has focused on the need as to why temporal graph theory is important for
IC-MANET. The analysis of DASW reveals that due to lack of time information,
routing protocol delivery ratio remains Epidemic like. Further, it confirms that an
abacus calculation for node degree is static, and confined to only RollerNet. Thus, our
attempt was to co-relate the existing theories and metrics of static graph with
temporal.
Numerous analyses with regard to information dissemination have been interpreted,
especially for improving the routing efficiency considering the dynamics of the real
time datasets. The observations that are made reveal that it can help for further
investigations into defining temporal graph and related metrics. To do so, we further
need to understand and analyze the existing shortest path, path length and centralities
with respect to time, and redefine them as temporal distance and temporal centrality.
This is being discussed in next chapter.
Chapter 4
Temporal AlgorithmCHAPTER 4. TEMPORAL ALGORITHM
4. Temporal Algorithm
39
4TEMPORAL ALGORITHM
As discussed in the previous chapter, that static graph and related metrics play a
significant role determining the temporal graphs; this chapter discusses temporal
metrics and algorithm. The temporal information divides into four categories:
timestamps, time-order, frequency and duration, used for defining temporal metrics.
Based on this, a temporal algorithm is proposed [73] and real traces have been
evaluated accordingly.
4.1 Temporal Graphs
Let’s consider the sequence of interaction as given in Table 4.1, representing the
traffic flow in network. From this, we can construct a temporal graph as shown in
Figure 4.1(a), and corresponding static graph in Figure 4.1(b). Here, interactions
between pairs of nodes define an edge or, equivalent, generated from the union of all
edges in the temporal graph[11].
4. Temporal Algorithm
40
Table 4.1 Interaction Sequence Between Nodes
(Source,Target) Timestamp Duration(A,B) 1 2(C,E) 2 1(E,F) 2 1(B,D) 3 1(C,D) 3 1
The first column defines pair of interacting nodes, the second column time, and the
third duration. Consider the path from node ‘A’ to ‘F’ using the static graph, there
exists a path from ‘A’ to ‘F’ via ‘B’,’D’,’C’,’E’. However, when taken into account of
the time information in temporal graph, it is seen that there is no path that satisfies
this route. It is due to the time-order; because the interaction between the sub-path
‘B’,’D’, ‘C’ and ‘C’, ‘E’, ‘F’ occurred in wrong time order to facilitate the path. Let’s
us define the concept of temporal graphs and path metrics.
(a) Temporal Graph (b) Static Graph
Figure 4.1 Temporal Graph
Definition 1 (Temporal Graph): In a given a real-world network since interaction
are traceable, starting at Tmin and ending at Tmax, the (undirected) temporal graphg (Tmin, Tmax) is defined as an ordered sequence of undirected graphs (G0, G1……GT-
1) where:
Gt = (Vt,Et), a 2-tuple consists of a set of nodes Vt and edges Et in the
window t;
4. Temporal Algorithm
41
there exists a link between node ‘i’ and node ‘j’ in Et if there is some link
in the real network between ‘i’ and ‘j’ during the time interval [(Tmin +(w× t)); (Tmin + (w × (t + 1)))];
T-1 = ((Tmax - Tmin) / w) = |g (Tmin, Tmax)| is the number of graphs in
the sequence;
w is the duration of each time window expressed in time units (e.g.,
seconds or hours); and |E| = ∑ |E| as the total number of edges across
all windows in the temporal graph.
This definition can be applied in the case of a directed temporal graph by means of a
sequence of directed graphs, where there exists a link from ‘i’ to ‘j’ in Et if there is a
contact from ‘i’ to ‘j’ during the time interval [(Tmin +(w × t));(Tmin + (w × (t + 1)))].
Simplifying Assumption
Let’s consider unweighted graphs since the datasets employed in this thesis
(RollerNet, INFOCOM’06) contain only binary contact information. However,
weighted temporal graphs would be a good case for future work. Further, we can refer
the set of nodes in a temporal graph as V = Vt, tϵ[0, T) and N = |V|.4.1.1 Temporal Metrics
Let’s focus on the definitions of temporal distance[13][16] and temporal
centrality[74].
Temporal Paths and Shortest Path Length
As we have highlighted in the Section 2.3, fundamental to the study of information
dissemination in networks is the concepts of paths and its lengths. However, the
shortest path length on static graphs returns the number of hops from a source to
destination; and does not retain temporal information[75], and hence cannot capture
the actual duration or speed of dissemination. Let’s redefine this metric as the shortest
4. Temporal Algorithm
42
temporal path length giving an indication of speed of the message delivery from a
source to destination. This is presented through a calculation of the temporal shortest
path length, and then proposing an algorithm to compute the temporal distance and
closeness centrality.
Definition 2 (Temporal Path): A temporal path, p = ((no , n1 …nn )starting at node ‘i’ = n0 and finishing at ‘j’ = nn, is defined over g (Tmin, Tmax) as a
sequence of h hops via a distinct node n at time window Wa where node ‘na’
passes the message if and only if there is an edge between na-1 and na at time window
Wa-1 ≤ Wa and 0 ≤ Wa < T, horizon parameter h, indicating the maximum number of
hops.
By referring to Figure 4.1(a), and calculating the temporal shortest path from node
‘A’ to ‘C’ with horizon h = 2, there is a temporal sub-path (A;B) at window 1 and
(B;D;C) at window 3. The horizon parameter is interpreted as the hops that a message
travels through the network and is directly related to the window size w, used to
model the network.
Definition 3 (Temporal Distance): For a given nodes ‘i’ and ‘j’, the shortest
temporal distance[16] is defined:d (T , T ) (4.1)as the shortest temporal path length, starting from time T , can be taken as the
number of time windows (or temporal hops h) which spreads information from node
‘i’ to ‘j’.
In case of temporarily disconnected node pairs (q,p), that is, information from ‘q’
does not reach to ‘p’, then set the temporal distance d = ∞. Then, temporal
distance d (T , T ) is computed in terms of number of time windows
i.e., d (T , T )= d (T , T ). Next, we propose an algorithm that describes
computing d (T , T ).
4. Temporal Algorithm
43
For each node pair of ‘i’ and ‘j’, d (T , T ) takes the average of all values. This
way, temporal distance is computed in a number of timestamps resulting temporal
distances in terms of time (i.g., in seconds). This is derived by multiplying it with
window size w. Eq. (4.2) giving average temporal distance between T and T :
L(T ,T )=ωN(N − 1) d (T , T ) (4.2)
In order to compute the temporal distance dij(T , T ) for a given time window,
let’s discuss and derive a methodology for calculating time window:
4.1.2 Time Window Calculation
Let’s assume that there are 6 nodes in a network and we build 6 by 6 adjacency
matrix, showing between each pairs (i, j): total contact time between (i, j) divided by
total number of contact occurrence with the same pair (i, j) during time (Tmin,Tmax).
For calculation of time window, refer Table 4.2 on a hypothetical datasets, where each
cell value represents the total contact time between node pairs (i, j) divided by the
total number of contact occurrences between the node pairs (i,j). For each node pairs
(i,j) compute a sum of all values, then returns the average meeting time per contact.
Table 4.2 Time Window Calculation
NodeID 1 2 3 4 5 6 ∑ .1 0/0 480/2 720/3 480/2 960/4 480/2 3120/132 500/2 0/0 750/3 250/1 1000/4 500/2 3000/123 735/3 490/2 0/0 490/2 735/3 245/1 2695/114 235/1 940/4 1175/5 0/0 705/3 470/2 3525/155 1300/5 260/1 780/3 520/2 0/0 1040/4 3900/156 510/2 510/2 255/1 1530/6 1275/5 0/0 4080/16∑T (T , T )∑N 2032082
Time Window Size 247.80
4. Temporal Algorithm
44
As shown in Table 4.2, the optimal value of time window should be greater than the
average meeting time. Because if time window <= average meeting time, then in most
of the time windows, number of contact occurrence is around one suggesting, that the
information cannot be diffused efficiently into the network. Now it is established that,
for effective information diffusion ,time window should be greater than∑ ( , )∑
i.e., time window = 300 seconds, resulting into the total number of time window =
(Tmax – Tmin) /w = (900 – 0) /300 = 3 timestamps.
Let’s find temporal distance d (t , t ) for the temporal graph as shown in Figure
4.1(a). Here, Tmin = 0 and Tmax = 900. Time window size = 300. Thus, there are 3 time
windows t1, t2 and t3.
4.1.3 Preconditions
Before starting calculation of temporal distance of each pair (i,j), initialize number of
empty lists equal to that of calculating number of time windows. When there is a
contact edge between (i,j) then the node pair (i,j) occurs. For each pair (i, j) i≠j, start
scanning timestamps from 1 to 3, and for each timestamp, add occurred node id into
the respective lists of timestamps.
Case 1
If i = = j then, return 0. Thus, temporal distance for (A, A) = (B, B) = (C, C) = (D, D)
= 0
Case 2
If both ‘i’ and ‘j’ occurs in same timestamp, then return (jth timestamp number – ith
timestamp number) or return (0). In Figure 4.1(a), node ‘A’ and node ‘B’ occurs in
same timestamp number 1, thus temporal distance between ‘A’ and ‘B’ is (B’s
timestamp – A’s timestamp) = (1-1) = 0
4. Temporal Algorithm
45
Case 3
If ‘i’ occurs earlier than ‘j’, then search the occurrences of ‘j’ in consecutive
timestamps by using other occurred nodes in the same timestamp, in which ‘i’ has
occurred. For each pairs (i,j) it may give more than one path in terms of required
timestamp, in such a case we have to select the shortest timestamp.
In Figure 4.1(a), for temporal distance (A,D), ‘A’ occurred in timestamp 1, and node
‘D’ occurred in timestamp 3. Also there is an intermediate node ‘B’, which is
common between ‘A’ and ‘D’. So, temporal distance (A,D) = (node D’s timestamp
number – node A’s timestamp number) = (3 – 1) = 2 timestamps.
Case 4
If ‘i’ occurs and ‘j’ does not during consecutive timestamps till Tmax, then the
temporal path between pairs (i,j) is not possible. So returns ∞.
In Figure 4.1 (a), for a temporal distance (D,E), node ‘D’ occurred in timestamp 3.
But there are no occurrences of ‘E’ in subsequent timestamps. So, temporal distance
(D, E) = ∞. Referring to Figure 4.1(a) presenting a detailed explanation per time
window and calculation of time stamps to load 6 by 6 matrix.
Time Window 1: There is a contact edge between pair (A,B). Since, A and B both
have occurred in the same time window, returns 0 for (A,B) and (B,A). Rest all node
pairs(A,A), (C,A), (D,A),(E,A) and (F,A) returns -1 (no connection).
Time Window 2: Here, contact edges are between the pairs (A,B), (B,A), (C,E),
(E,C), (E,F), (B,B) and (F,E), and all of them have occurred in the same time window,
hence returns 0. For pair (D,F) there is no connection and hence, returns -1( no
connection).
Time Window 3: The contact edges for node pairs (D, B) and (D, C) have occurred
in the same time window hence returns 0. Rest all the pair returns 1 as per case 3 in
their respective time windows. Thus, Temporal Distance Matrix =
4. Temporal Algorithm
46
[ [ 0, 0, 1, 1 -1, -1 ],
[ 0, 0, 1, 1, -1, -1 ],
[ -1, 1, 0, 1, 0, 0 ],
[ -1, 0, 0, 0, -1, -1 ],
[ -1, 1, 0, 1, 0, 0 ],
[ -1, 1, 0, 1, 0, 0 ] ]
In continuation of the examples presented in Figure 4.1(a) and successive
computation of temporal distance matrix as above, the sum of non-negative values of
matrix = 10.
Now, to calculate the average temporal distance metric: 300 (10/ (6) (5)) =3000/ (30)
=100. i.e., it takes average 100 seconds to reach from ‘i’ to ‘j’.
4.2 Temporal Algorithm
1. Input Source,Target, Tmin,Tmax Time Window
2. Time Window Equation:w(Timewindow) > ∑T (T , T )∑N (4.3)Where, ∑T (T , T ) = Total contact time between all pairs and∑N = Total occurrences of all pairs (i, j)
3. Number of time frames = (Tmax – Tmin) / Time window
4. Initialize number of empty list equal to the number of time frames. Each list
shows node ids whose contact occurred in a respective time frame.
5. Read the dataset and perform scan for node contact in different time frames and
generate a distance matrix.
6. Per contact frame, fill up the array with node ids in contact.
7. Compute the temporal distance as:
a. If source and target ids are in the same list, return (target time frame
number – source time frame number) as temporal distance.
4. Temporal Algorithm
47
b. Otherwise, search source and target in different time frames. If the
source time frame < target time frame, then return (target time frame
number – source time frame number) as temporal distance.
c. In case of repeated occurrence of the (source, target) set Tmin= last
target occurred +1 timestamp and repeat steps a and b.
8. Take average values of all pairs (source, target) of temporal distance.
9. Repeat steps 4,5,6 and 7 for all pairs (source, target) and generate matrix. -1
(minus one) indicating no edge between a pair of nodes.
The follow-up section explores the importance of nodes for information dissemination
based on temporal centralities.
4.2.1 Temporal Betweenness Centrality
Betweenness is used to discover nodes that are critical for mediating information
flow[74]. To identify these mediating nodes as described in Section 3.2.2, the static
betweenness centrality of node ‘i’ is the shortest path between all pairs of nodes that
passes through ‘i’. This proportion is important as it gives a higher weight to nodes
which facilitates paths where there are no alternatives.
To capture the notion of temporal betweenness it is important to take into account not
only the proportion of shortest paths which passes through a node, but also the length
of time for which a node along the shortest path retains a piece of information before
forwarding it to the next node. For example, consider the 2-hop shortest temporal path
from node P to R, (P;Q;R) as shown in Figure 4.2.
Figure 4.2 Betweenness Centrality
4. Temporal Algorithm
48
In terms of time, this path is represented as (P;Q;Q;Q;Q;R) since message resides in
node Q for 4 time windows, by removing this, will have a greater impact in disrupting
the network, as shown in Figure 4.3
Figure 4.3 Betweenness Centrality by Consideration of Time Duration
For a given time window T, the temporal betweenness centrality of node ‘i’ is defined
as: CiB(t) = 1(N− 1)(N − 2) U(i, t, j, k)Sjkh (4.4)k ∈Vk≠i,k≠jj ≤ Vj≠ iwhere,U(i, t, j, k) returns number of shortest paths from ′j′ to ′k′, which node ′i′ is holdingthe message at time window tSjkh indicates number of shortest temporal paths between ‘j’and ‘k’. Finally, the
average temporal betweenness value across all time windows for each node ‘i’ is:
CiB(t) = 1W CiB ((t × w)+ tmin )Wt=1 (4.5)
4.2.2 Temporal Closeness Centrality
As known two nodes of a static graph are close to each other if their geodesic distance
is small. The same definition is applied to temporal closeness centrality by using the
4. Temporal Algorithm
49
temporal shortest path length to measure the speed at which a source node can deliver
a message to all other nodes.
Given the shortest temporal distance dij(Tmin,Tmax), temporal closeness centrality[12]
is expressed as: C = ( ) ∑ d ,∈ and the nodes having shorter temporal
distance to the other nodes can be considered more central.
The temporal closeness centrality scenario as presented in Figure 4.4 (a) to (d), the
total simulation time 12 is divided into 4 equal time windows (t1, t2, t3,t4 ), each of
duration 3.
Figure 4.4 Temporal Closeness Centrality
For t1 time window, closeness centrality of node A is:
C = ( ) ∑ d ,∈ = ∗( ) (1 + 1 + 1 + 2 + 2 + 2) = (9) = 0.5 (4.6)Similarly, it is calculated for all nodes in t1, t2, t3 and t4 time.frame as shown in Table
4.3. The average closeness centrality per time window (AVG) is the average of the
closeness centrality of all nodes in the same time window.
4. Temporal Algorithm
50
Table 4.3 Temporal Centrality Calculations
Node Closeness Centrality
A 0.5B 0.5C 0.444D 0.667E 0.667F 0.722G 0.722
AVG 0.603
Node Closeness Centrality
A 0.933C 0.533D 0.667E 0.533F 0.8G 0.8
AVG 0.711
Node Closeness Centrality
B 0.556C 0.444D 0.333E 0.444
AVG 0.444
Node Closeness Centrality
A 0.333B 0.533D 0.333E 0.533F 0G 0.4
AVG 0.355
Average Closeness Centrality (ACC)
Every node has closeness centrality at its time window (AVG), based on that, the
Average Closeness Centrality (ACC) is calculated as:
ACC = (0.603 + 0.711 + 0.444 + 0.355)4 = 2.1134 = 0.528 (4.7)4.3 Application of Temporal Algorithm
Let’s apply temporal distance and temporal centrality to real networks that exhibit
time information. This algorithm13 is implemented using networkX package of
python. Figure 4.5 outlines the methodology, approach of temporal metrics evaluation
and its utilization. This demonstrates the applicability of metrics on RollerNet,
13 Temporal Algorithm is implemented using networkX package of Python. Refer the script inAppendix –C.3 for further study.
t1 t2
t3 t4
4. Temporal Algorithm
51
INFOCOM’06 and RWP. The next chapter discusses more about the applicability of
temporal metrics for efficient routing.
Figure 4.5 Temporal Metrics Evaluation and Utilization Scenario
4.3.1 Experimental Dataset and Common Format
A range of empirical networks have been discussed in Section 2.3, with regard to
temporal information and datasets like RollerNet, INFOCOM’06 and RWP14. Table
4.4 narrates the information about duration, start and end times, number of contacts
and nodes for experimental datasets.
14 RWP is random family mobility model. It has been included to study and compare temporalinformation that of the real traces.
4. Temporal Algorithm
51
INFOCOM’06 and RWP. The next chapter discusses more about the applicability of
temporal metrics for efficient routing.
Figure 4.5 Temporal Metrics Evaluation and Utilization Scenario
4.3.1 Experimental Dataset and Common Format
A range of empirical networks have been discussed in Section 2.3, with regard to
temporal information and datasets like RollerNet, INFOCOM’06 and RWP14. Table
4.4 narrates the information about duration, start and end times, number of contacts
and nodes for experimental datasets.
14 RWP is random family mobility model. It has been included to study and compare temporalinformation that of the real traces.
4. Temporal Algorithm
51
INFOCOM’06 and RWP. The next chapter discusses more about the applicability of
temporal metrics for efficient routing.
Figure 4.5 Temporal Metrics Evaluation and Utilization Scenario
4.3.1 Experimental Dataset and Common Format
A range of empirical networks have been discussed in Section 2.3, with regard to
temporal information and datasets like RollerNet, INFOCOM’06 and RWP14. Table
4.4 narrates the information about duration, start and end times, number of contacts
and nodes for experimental datasets.
14 RWP is random family mobility model. It has been included to study and compare temporalinformation that of the real traces.
4. Temporal Algorithm
52
Table 4.4 Experimental Datasets15
RollerNet RWP_63 INFOCOM’06 RWP_98
Start Date 2/2/2009 NA 13/03/2005 NA
Duration 0.12 day 0.12 day 4 days 4 days
Start-EndTime
Day 1: 0-3096
( 51.6 min)
0- 3096 Day 1: 61260 - 86400
(6.98 hours)
Day 2: 86400 - 172800
(24 hours)
Day 3: 172800 - 259200
(24 hours)
Day 4: 259200 –345600
(24 hours)
0-342915
Numberof Nodes
63 63 98 98
Contacts 80824 576 118875
( of all four days)
4412929
Common Dataset Format
In order to carry out time window calculation for any real trace or synthetic dataset,
we have set the common format as in Table 4.5. The customized16 scripts are written
by taking care of converting datasets into desired format.
Table 4.5 Common Format for Experimental Datasets
SourceNode ID
DestinationNode ID
ConnectionUptime
ConnectionDowntime
OccurrenceCount
IntercontactTime
1 3 51293 51293 1 01 3 60603 60603 2 93101 3 62363 62363 3 17601 3 79649 79649 4 17286
15 Experimental datasets includes RollerNet, INFOCOM ’06 and RWP.Further details regardinglocation, duration, participants and devices used refer Appendix B.16 Refer Appendix C.1 and C.2 for script that converts dataset into common format and generatesconnectivity report.
4. Temporal Algorithm
53
4.3.2 Time Window Size Calculation for Experimental Datasets
For time window calculations, the methodology as discussed in the Section 4.2.1, the
calculation is presented in Table 4.6 .
Table 4.6 Time Window Calculations for RollerNet, INFOCOM’06 and RWP
Dataset Details Tmin
(sec.)
Tmax
(sec.)
TotalNode
s
Numberof
Connection
Numberof
Time-stamp
s
Timewindow (w)(sec.)
INFOCOM'06Day 1: 61260 -
86400 (6.98 hours)61260 86400 96 178695 8 3240
Day 2: 86400 -172800 (24 hours)
86400 172800 98 585414 26 3240
Day 3: 172800 -259200 (24 hours)
172800 259200 93 378624 27 3240
Day 4: 259200 -345600 (24 hours)
259200 345600 83 9227 27 3240
RollerNetDay 1: 0-3096
(51.6 min)0 3105 62 2711107 207 15
RWP_630- 3096 0 3124 25 576 44 71
RWP_980- 342915 0 342936 98 4412929 2598 132
The above table indicates the values of Tmin, Tmax, nodes, number of connections,
timestamps and window size which is used as input by temporal algorithm to compute
temporal distance and centrality.
Next, the average temporal distance calculation is presented. Temporal Algorithm
builds temporal distance matrix and performs the summation of non-negative values
of matrix. Using equation 4.2, it computes the average temporal distance.
4. Temporal Algorithm
54
4.3.3 Temporal Distance Evaluation
To compare between temporal and static metrics, the results are calculated for the
INFOCOM’06, RollerNet and RWP as shown in Table 4.7. Since the paths in static
graphs ignore duration of contacts, inter-contact time, recurrent contacts, and time
ordering of contacts, it overestimates the number of connected node pairs and
underestimates the path lengths.
Table 4.7 Temporal Distance for RollerNet, INFOCOM’06 and RWP
Dataset Details Tmin(sec.)
Tmax(sec.)
Numberof
Time-stamp
s
Timewindow
(w)(sec.)
StaticDistanc
e
AverageTempor
alDistanc
e
INFOCOM'06Day 1: 61260 -86400 (6.98 hours)
61260 86400 8 32401.56 0.25
Day 2: 86400 -172800 (24 hours)
86400 172800 26 32401.23 0.51
Day 3: 172800 -259200 (24 hours)
172800 259200 27 32401.3 0.23
Day 4: 259200 -345600 (24 hours)
259200 345600 27 32401.3 1.47
RollerNetDay 1: 0-3096( 51.6 min)
0 3105 207 151.81 0.31
RWP_63170- 3096 0 3124 44 71 1.12 1.05
RWP_980- 342915 0 342936 2598 132 3.64 0.32
The above temporal metric gives us a better understanding of the network with
respect to the temporal dimension providing an accurate measure of delay for
information diffusion process as it was hidden in static.
17 For RWP_63 temporal distance values are computed and presented in Table D 1 of Appendix D
4. Temporal Algorithm
55
4.3.3 Temporal Closeness Centrality Evaluation
Table 4.8 presents different centrality static values for experimental datasets. This is
indicated by two tuple (t1,t2) where, t1 = Node id and t2= Centrality value. It is
computed without considering the time information.
Table 4.8 Closeness Centrality for RollerNet, INFOCOM’06 and RWP
Dataset Details Diameter DegreeCentrality
BetweenesCentrality
ClosenessCentrality
INFOCOM'06Day 1: 61260 - 86400(6.98 hours)
4 (27, 0.81) (16, 0.04) (47, 0.83)
Day 2: 86400 - 172800(24 hours)
3 (56, 0.98) (16, 0.02) (56, 0.98)
Day 3: 172800 - 259200(24 hours)
3 (51, 0.95) (51, 0.01) (51, 0.95)
Day 4: 259200 - 345600(24 hours)
3 (44, 0.77) (30, 0.05) (44, 0.80)
RWP_980- 342915 2 (17, 0.99) (17, 0.09) (17, 0.99)
RollerNetDay 1: 0-3096( 51.6 min)
2 (51, 0.1) (51, 0.004) (51, 1.0)
RWP_630- 3096 6 (14, 0.29) (14, 0.65) (15, 0.43)
Next, the value of temporal closeness centrality are computed for INFOCOM’06,
RollerNet and RWP_63, and the table column shows a number of time stamps, node
id and its centrality value per timestamp in Appendix –D.
It is observed that the static value of centrality only considers node ids with the
highest centrality values, wherein, temporal closeness centrality values shows the
node id with centrality values per timestamp. Thus, closeness centrality values may
vary from timestamp to timestamp.
4. Temporal Algorithm
56
4.4 Observations
Optimal time window size varies as per number of connections between nodes,
number of nodes and total duration of Tmin,Tmax. Keeping the value lower than
the derived through customized script may result in overlooking connection and
keeping high resulting in wastage of network resources.
The synthetic dataset values for temporal distance are poorer than the real trace,
due to its characteristics moving towards centre at random. Average temporal
distance values of real trace analysis enables taking realistic routing decisions
accurately (Findings are published and presented in WORLDCOMP’13)
Different centrality values help in identifying the important nodes. Such nodes
assist in the efficient information dissemination process. Further, closeness
centrality values per timestamps help in determining more consistent nodes and
its average using network density checks.
Referring to the readings of RollerNet and RWP it reveals in RWP model the
node movements are random and hence, number of contacts and timestamps are
less, resulting in lower average temporal distance value. It is seen that most of
the time the nodes are moving around the centre due to which the diameter,
degree centrality, betweenness and closeness values are higher. These values
clearly indicating the reasons (described above) behind limited uses of synthetic
models.
On the other hand, RollerNet has comparatively higher contacts, and higher
number of timestamps resulting better connectivity. Therefore, for efficient
information dissemination these characteristics are being used by routing
engines. (Findings are published and presented in WORLDCOMP’13)
4. Temporal Algorithm
57
Chapter Summary
It reveals that the node mobility plays a vital role for efficient diffusion of information
in challenged environments. And while doing so one cannot ignore to understand the
movement patterns and related properties, such as time order, frequency, contact
duration, inter contact time etc. These dynamic properties of connection are first
analyzed and understood by using time varying matrices like temporal distance and
temporal centrality.
The algorithm has been generalized and designed for evaluating temporal metrics
from any synthetic and real trace. Because such frameworks help in computing
number of time frames and size of time windows, which in turn calculate temporal
properties. These properties are very useful in understanding the dynamics of network
especially in decision makings be it forwarding or replication.
Chapter 5
Adaptive RoutingCHAPTER 5. ADAPTIVE ROUTING
5. Adaptive Routing
58
5ADAPTIVE ROUTING
The IC-MANET routing is classified as Forwarding and Replication technique. In
Forwarding single message copy is used whereas in Replication it is multiple.
Through this chapter we propose a routing that integrates both techniques. It checks
network density that determines the sparse and dense in a given time interval. If the
network is dense routing engine calls for an Encounter Based Forwarding[53], else, it
invokes Two Period Spray & Wait[54]. Thus, we name it Adaptive Routing (AR) that
toggles between Forwarding and Replication.
As discussed in Section 3.1 DASW performs the lookup for number of message
copies into abacus. By applying the concept of temporal distance in DASW the results
are seen better which can be used with any real traces like INFOCOM’05.
5. Adaptive Routing
59
5.1 Analyzing Modified DASW
It is observed in Section 3.1 that the existing DASW[63] is adaptive18 in nature, but
the limitations are:
1. That the technique performs static lookup for number of message copies,
2. the number of message copies are fixed,
3. and it works only for RollerNet.
We have applied Temporal Algorithm as proof of concept to modify the DASW and
overcome the challenges mentioned above. Now, the modified DASW replicates the
message copies based on each node’s neighborhood index per time window, thus,
ensuring that the limited numbers of message copies are spread.
Next, the performance of modified DASW is evaluated and compared with Epidemic
Routing using INFOCOM’05 for delivery ratio. It is seen that the overall result is still
poor, but temporal characterization works for modified DASW.
5.1.1 Performance Evaluation of Modified DASW usingINFOCOM’05
Figure 5.1 Message Event Interval vs Delivery Ratio for Modified DASW usingINFOCOM’05
18 As per the need to achieve average delay and node degree, it looks up to abacus for spreading theappropriate number of message copies.
00.020.040.060.08
5-10 10-15 15-20 20-25 25-30Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
DASW
Epidemic
5. Adaptive Routing
60
It is observed in Figure 5.1 that overall delivery ratio is quite poor for modified
DASW and Epidemic. This is due to very low neighborhood index (0.051). Thus,
modified DASW could replicate only a few copies. But, delivery ratio is seen slightly
better than Epidemic. Earlier DASW and Epidemic were behaving the same way and
spreading around 128 message copies. Now, it is 0.051.
5.2 Forwarding
Encounter based technique is used in forwarding phase. Here each node maintains
encounter value (EV) and stores the frequently encountered node ids in summary
vector F[ ]. To track the node’s encountered rate, each node maintains two pieces of
local information: EV and Current Window Counter (CWC). EV represents the
node’s past rate of encounters as an exponentially weighted moving average, while
CWC obtains information about the number of encounters in the current time interval.
EV is periodically updated taking into account of the most recent CWC and the
updated EV is computed as Eq.(5.1) :
EV ← α・CWC + (1 − α)・EV (5.1)
The exponentially weighted moving average places emphasis proportional to α on the
most recent complete CWC. Updating CWC is straightforward, for every encounter,
CWC is incremented. When the current window updates interval expires, the EV is
updated and the CWC is reset to zero.
Suppose, node ‘n’ encounters the node ‘m’, if ‘n’ has a message for ‘m’ then it sends
the message to ‘m’. After that if ‘n’ has a message for the destination id which is
stored in summary vector F[] of ‘m’ then ‘n’ forwards the message to ‘m’. Otherwise,
if the encounter rate EV of ‘m’ is greater than that of the encounter rate of ‘n’ then it
forwards the message to ‘m’.
5. Adaptive Routing
61
5.2.1 Algorithm: Encounter based Forwarding
F [ ] = vector of frequently encountered nodes
V [ ] = message vector which carries a destination id of the message
Wi= current window update interval
Upon reception of a Hello message h from node m do
ifnewNeighbour(m) == true
ifmsgQueue.hasMsgsForDest(m) == true
deliverMsgs(m)
updateEV()
for all destinations d ∈n.V[]do
if d∈m.f[]
forward message to node m
else m.EV>n.EV
forward message to node m
updateEV()
if time ≥ nextUpdate then
EV ← α ・CWC + (1 − α) ・EV
CWC ← 0
nextUpdate← time +Wi
end if
5.3 Replication
It is based on Two Period Spray & Wait[54], the algorithm starts with spraying fewer
message copies than the decided one’s. It waits for a certain period of time to see if
the message is delivered. When the delivery does not happen, the algorithm sprays
additional copies of a message and waits for the delivery.
5. Adaptive Routing
62
5.3.1 Algorithm: Two Period Spray & Wait
Sprays L1 copies to the network at the beginning of execution and additional L2-L1
copies at next time interval.
TwoPeriodsw(L, α, td )
optcost=L;
for each 0 < L1 < L do
L2floor = max[ L+1, L1+(α L1 td (L- L1))];
for L2 = L2floor , L2floor +1 do
if c2(L1, L2 ) < opt_cost then
opt_cost = c2(L1, L2 ); opt_cts = [L1, L2]
end if
end for
end for
return opt_cts
Where,
td = Time deadline
L = Number of message copies
α = 0.0015 x number of copies
L1= Number of message copies
to spray in the first time period
L2 = Number of message copies
to spray in the second time period
c2 = Cost as number of copies used
per message
5.4 Algorithm: Adaptive Routing
Input: Tmin, Tmax, Average Temporal Distance
Output: Delivery Ratio, Overhead Ratio, Number of Dropped Messages, Average
Latency of Delivered Message
Procedure:
1. Read the values of Tmin, Tmax and average temporal distance from
experimental datasets.
2. Compute the threshold for the Average temporal distance. At the end of
each timestamp, calculate the gain of average temporal distance comparing
to previous timestamp.
5. Adaptive Routing
63
3. Threshold = average of all the gain calculated at the end of each
timestamp.
4. If average temporal distance < threshold, then do call Encounter Based
Forwarding otherwise call Two Period based Spray & Wait.
5. Generate messagstat report.
6. Plot the comparison chart of Message Interval with Delivery Ratio,
Overhead Ratio, Number of Dropped Messages and Average Latency of
Delivered Messages.
5.4.1 Average Temporal Distance Gain
It is calculated as Average Gain of Temporal Distance = ∑ (Difference of all
temporal distance for current and previous timestamps) divided by total number of
timestamps.
Consider values of temporal distance available for timestamp 1 to 5 as presented in
Table 5.1. Temporal distance gained is computed as the difference between temporal
distance of current timestamp and previous, e.g., Temporal Gain for timestamp 3 =
2.76 – 0.02 = 2.74. Thus, gain in temporal distance indicates that there are changes in
network conditions and it is used as threshold for determining the network density by
AR.
Table 5.1 Average Temporal Distance Gain Calculations
TimestampNumber
TemporalDistance
TemporalDistance
GainRemarks
1 0 0
2 0.02 0.02 (0.02-0)
3 2.76 2.74 (2.76-0.02)
4 3.26 0.5 (3.26-2.76)
5 3.67 0.41 (3.67-3.26)
Average TemporalDistance Gain A=
0.734
5. Adaptive Routing
64
5.5 Temporal Closeness Centrality based Approach
Generally closeness centrality is used for identifying important nodes; here it is used
for routing decisions. When two nodes encounter if the target becomes destination,
then message is directly delivered. Otherwise, algorithm compares its Average
closeness centrality per time window (AVG) with Average Closeness Centrality
(ACC) per time frame. If AVG > ACC then network is dense and routing engine
decides to perform Encounter Based Forwarding else it performs Two Period Spray &
Wait.
Input: Tmin, Tmax, Temporal Closeness Centrality
Output: Delivery ratio, Overhead Ratio, Number of Dropped Messages, Average
Latency of Delivered message.
Procedure:
1. Upon the encounter of two nodes
2. If the encountered node is destination node, then
3. Transfer message to encountered node
Else
4. Compare AVG and ACC
5. If AVG > ACC, then
6. Perform Encounter Based Forwarding
Else
7. Perform Two Period Spray & Wait.
Chapter Summary
It is seen that by using the temporal properties in modified DASW, the delivery ratio
improves but not significantly. It enhances the capability of the modified DASW to
operate with any real trace, and based on node’s neighborhood index, message copies
are created dynamically.
5. Adaptive Routing
65
Temporal distance and temporal closeness centrality is applied for network density
checks and based on that Adaptive Routing uses average temporal gain to adapt
Forwarding or Replication technique. For efficient message delivery, the Encounter
Based Routing is modified for using in Forwarding phase, in Replication phase Two
Period Spray & Wait is used.
The next chapter discusses about simulation set-up and performance analysis of AR.
Chapter 6
Performance EvaluationCHAPTER 6. PERFORMANCE EVALUATION
6. Performance Evaluation
66
6CHAPTER 6. PERFORMANCE EVALUATION
AR is implemented by using ONE simulator. Its performance is compared with Spray
& Wait and PRoPHeT. First, the simulation environment set-up is briefed, followed
by a list of input parameters such as assumptions, datasets and performance metrics.
Simulation is carried out:
1. by using temporal distance and temporal centrality
2. with real traces (INFOCOM’06, RollerNet) and Synthetic (RWP) datasets.
Here, the objective is to measure delivery ratio, overhead ratio, number of dropped
messages and average latency of delivered messages. It confirms our findings that the
evaluation of temporal properties plays a key role for efficient transmission and
reception of message in IC-MANET. Further, it reveals the inherent characteristics of
synthetic datasets and the significance of real traces.
6. Performance Evaluation
67
6.1 Evaluation Approach based on Virtual Test Bench
We have used ONE (Opportunistic Network Environment) [17] simulator for
implementation of AR and experimentation. This simulator is discrete event based
and open source written in JAVA. It provides packages, libraries, routing protocols,
mobility models, reports, energy models and support for importing real traces. That is
why researchers have preferred ONE over NS-2 and OMNeT++ for DTN.
The simulation scenario consists of 4500 × 3400 m area for experimental datasets, 98
nodes are participating for INFOCOM’06 and RWP_98, 63 nodes are participating for
RollerNet and RWP_63. All nodes are moving with speed of 0.5 – 1.5 m/s and with
waiting time of 0 – 120 seconds. Two kinds of devices are used in the experiments,
one is Bluetooth device, whose transmission range is 10 m and speed is 2 Mbps. The
other is 802.11b WLAN device with a transmission range of 30 m and speed is 4.5
Mbps. The buffer of these mobile devices is upto 5 MB. When buffer is full, new
messages are not accepted by the node till some old messages are deleted from the
buffer. The duration is 342915 simulation seconds for INFOCOM’06 and RWP_98.
For Rollernet and RWP_63 the duration is 3096 simulation seconds.
6.1.1 Assumptions
It is assumed that the network is secured, transmission is error free and nodes have
infinite energy.
6.1.2 Datasets
Real and Synthetic datasets are used for simulation. As discussed in Section 4.3.1
INFOCOM’06 and RollerNet are used. For synthetic dataset RWP_YY is used,
where, YY denotes the number of nodes. Table B.1 of Appendix B discusses in detail
6. Performance Evaluation
68
about the name, location, duration, participants and address IDs for experimental
datasets. First temporal characteristics of real traces and synthetic datasets are
evaluated as presented in Table 4.6 and Table D1(a) to (k). Real traces are converted
into ONE compatible format by customized script and then imported for simulation.
RWP dataset is generated by using ONE simulator.
6.1.3 Performance Metrics
Delivery Ratio: It is the measure of a fraction of created packets delivered to
the destination. This is the ratio of the total number of packets that are delivered to the
total number of packets that are created.
Overhead Ratio: The overhead ratio is calculated using the following equation.
Overhead Ratio = Number of relayed messages − Number of delivered messagesNumber of delivered messagesHere, the term relayed messages refers to the messages that have been forwarded
from the source to an intermediate node. This number is a measure for the number of
packets or copies of packets that have been inducted. The number of delivered
messages refers to the total number of created packets that are successfully delivered.
Average latency of delivered message is measured and number of message dropped is
due to TTL expiry or buffer overflow.
6.1.4 Input Parameter Settings in ONE Simulator
Input parameter’ setting for ONE simulator is being summarized in three tables. Table
6.1 shows the original dataset values obtained from CRAWDAD, e.g., in real trace of
RollerNet, 63 nodes participated and for INFOCOM’06 it is 98.
6. Performance Evaluation
69
Table 6.1 ONE Settings for INFOCOM’06 and RollerNet
Dataset INFOCOM’06 RollerNetScenario.simulateConnections False FalseScenario.updateInterval 1 1Scenario.endTime 342915 3096Scenario.nrofHostGroups 1 1Group.nodeLocation 10,10 10,10Group.bufferSize 5M 5MGroup.movementModel StationaryMovement StationaryMovementGroup.router AR ARGroup.nrofHosts 98 63Events.nrof 2 2Events1.class MessageEventGenerator MessageEventGeneratorEvents1.interval(Message generation timeinterval at each node)
5-10, 10-15, 15-20,20-25,25-30,30-35
5-10, 10-15, 15-20,20-25,25-30,30-35
Events1.hosts 0,97 0,62Events2.class ExternalEventsQueue ExternalEventsQueue
Events2.filePathhaggle6-INFOCOM’06.csv
RollerNet.dat
Similarly, for comparison Table 6.2 presents RollerNet vs RWP_63 and Table 6.3
presents INFOCOM’06 vs RWP_98 to set values of synthetic datasets (RWP_63,
RWP_98).
Table 6.2 ONE settings for RollerNet and RWP_63
Dataset RollerNet RWP_63Scenario.simulateConnections False FalseScenario.updateInterval 1 1Scenario.endTime 3096 3096Scenario.nrofHostGroups 1 1Group.bufferSize 5M 5MGroup.nodeLocation 10,10 10,10Group.movementModel StationaryMovement StationaryMovementGroup.router AR ARGroup.nrofHosts 63 63Events.nrof 2 2Events1.class MessageEventGenerator MessageEventGeneratorEvents1.interval 5-10, 10-15, 15-20,20-25 5-10, 10-15, 15-20,20-25Events1.hosts 0,62 0,62Events2.class ExternalEventsQueue ExternalEventsQueueEvents2.filePath RollerNet.dat Synthetic_63.datUpdateInterval 15 71
6. Performance Evaluation
70
Table 6.3 ONE settings for INFOCOM’06 and RWP_98
Dataset INFOCOM’06 RWP_98Scenario.simulateConnections False FalseScenario.updateInterval 1 1Scenario.endTime 342915 342915Scenario.nrofHostGroups 1 1Group.bufferSize 5M 5MGroup.nodeLocation 10,10 10,10Group.movementModel StationaryMovement StationaryMovementGroup.router AR ARGroup.nrofHosts 98 98Events.nrof 2 2Events1.class MessageEventGenerator MessageEventGeneratorEvents1.interval 5-10, 10-15, 15-20,20-25 5-10, 10-15, 15-20,20-25Events1.hosts 0,97 0,97Events2.class ExternalEventsQueue ExternalEventsQueue
Events2.filePathhaggle6-INFOCOM’06.csv
Synthetic.dat
UpdateInterval 3240 132
6.2 Graphical Analysis and Representation of the Findings
Simulation scenarios are classified into three groups: Group-1 takes care of evaluating
AR with real and synthetic using experimental datasets, Group-2 discusses about
comparison of AR with Spray & Wait and PRoPHeT using temporal distance, Group-
3 describes about the comparison of AR using temporal closeness centrality.
In Chapter 4 we have discussed Temporal Characterization Algorithm, in which
Section 4.3 presents evaluation of temporal distance and temporal closeness centrality
for RollerNet, INFOCOM and RWP. The algorithm is integrated with AR in ONE
simulator (as shown in Figure 4.5) to evaluate any real and or synthetic traces, and
accordingly it takes routing decisions.
Group-1: Evaluation of AR with Real and Synthetic Experimental Datasets
6. Performance Evaluation
71
6.2.1 Delivery Ratio Comparison of AR with RollerNet vs RWP_63
RWP_63 dataset has less connections compared to RollerNet. This is due to random
movement of the nodes and its tendency to move towards the centre resulting poor
temporal distance gain for RWP_63. Thus, the network is evaluated sparse. Therefore,
AR calls for Replication rather than Forwarding, and delivery ratio becomes lower for
RWP_63 as shown in Figure 6.1.
Figure 6.1 Delivery Ratio Comparison of AR with RollerNet vs RWP_63
6.2.2 Overhead Ratio Comparison of AR with RollerNet vs RWP_63
Figure 6.2 Overhead Ratio Comparison of AR with RollerNet vs RWP_63
0
0.2
0.4
0.6
0.8
1
5-10 10-15 15-20 20-25 25-30 30-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
RollerNet
RWP_63
01020304050607080
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
RollerNet
RWP_63
6. Performance Evaluation
72
During temporal evaluation the RollerNet has large connections. This is due to large
participation of tourists in Rollerblader tour. Thus, it is found that the temporal
distance gain for RollerNet is higher and network is observed dense. Hence, AR calls
for more Forwarding than Replication, and the overhead ratio is lower as shown in
Figure 6.2.
6.2.3 Number of Dropped Messages Comparison of AR withRollerNet vs RWP_63
Figure 6.3 Message Drop Comparison of AR with RollerNet vs RWP_63
AR with RollerNet has better temporal distance gain and has executed forwarding
technique more number of times than Replication. In Forwarding, number of
transmissions is less and single copy of a message exists in network. Thus, RollerNet
has lower value for number of message dropped compared to RWP_63 as shown in
Figure 6.3
6.2.4 Average Latency of Delivered Message Comparison of AR withRollerNet vs RWP_63
0
1000
2000
3000
4000
5000
6000
5-10 10-15 15-20 20-25 25-30 30-35
Num
ber o
f Dro
pped
Mes
sage
s
Message Event Interval
Message Event Interval vs Number of Dropped Messages
RollerNet
RWP_63
6. Performance Evaluation
73
It is observed that the temporal distance gain for RollerNet is higher and the network
is dense, it is due to higher number of connections. Hence, AR calls for more
Forwarding than Replication. In Forwarding, number of transmissions is less and
single copy of a message exists in network. Thus, RollerNet has lower latency values
of delivered messages compared to RWP_63 as shown in Figure 6.4
Figure 6.4 Average Delivered Message Latency Comparison of AR withRollerNet vs RWP_63
6.2.5 Delivery Ratio Comparison of AR with INFOCOM’06 vsRWP_98
Figure 6.5 Delivery Ratio Comparison of AR with INFOCOM’06 vs RWP_98
0200400600800
100012001400
5-10 10-15 15-20 20-25 25-30 30-35
Ave
rage
Lat
ency
of D
eliv
ered
Mes
sage
s
Message Event Interval
Message Event Interval vs Average Latency of Delivered Messages
RollerNet
RWP_63
0
0.05
0.1
0.15
0.2
0.25
5-10 10-15 15-20 20-25 25-30 30-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
INFOCOM
RWP_98
6. Performance Evaluation
74
There is a marginal difference in delivery ratio of INFOCOM’06 and RWP_98 but, as
observed the overall delivery ratio is poor as shown in Figure 6.5. This is because the
INFOCOM’06 has a smaller number of timestamps and larger time window size
compared to RWP_98. Hence, AR with INFOCOM’06 executes Forwarding and
Replication, almost same number of times. RWP_98 has a large number of
timestamps and short time window size. Thus, RWP_98 executes forwarding more
number of times upto certain time interval resulting in poor delivery ratio compared to
INFOCOM’06.
6.2.6 Overhead Ratio Comparison of AR with INFOCOM’06 vsRWP_98
Figure 6.6 Overhead Ratio Comparison of AR with INFOCOM’06 vs RWP_98
In RWP_98 the number of connections is more compared to INFOCOM’06. Thus, the
network is dense and AR frequently executes Forwarding resulting in less overhead
compared to INFOCOM’06 as shown in Figure 6.6. Further, the number of
timestamps is less with higher time window size resulting in additional dropping of
the messages in buffer queues.
0
20
40
60
80
100
120
140
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
INFOCOM
RWP_98
6. Performance Evaluation
75
6.2.7 Average Latency of Delivered Messages Comparison of AR forINFOCOM’06 vs RWP_98
Figure 6.7 Average Delivered Message Latency Comparison of AR withINFOCOM’06 vs RWP_98
INFOCOM’06 has a smaller number of timestamps and larger time window size
compared to RWP_98. AR with RWP_98 executes Replication more than
INFOCOM’06. Thus, incurred larger message drop and higher averge latency
compared to INFOCOM’06 as shown in Figure 6.7.
Group-2: Comparison of AR using temporal distance with Spray & Wait andPRoPHeT
6.2.8 Delivery Ratio Comparison of AR with Spray & Wait andPRoPHeT using INFOCOM’06 and RWP_98
The delivery ratio in AR is better than that of Spray & Wait and PRoPHeT as
observed in Figure 6.8(a). But, overall value is still poor. This is because
INFOCOM’06 has a smaller number of timestamps and larger time window size
compared to RWP_98. Hence, AR with INFOCOM’06 executes Forwarding and
Replication equal number of times. Due to lower value of node degree (0.20) delivery
ratio of Spray & Wait and PRoPHeT is found below AR.
0
1000
2000
3000
4000
5000
6000
5-10 10-15 15-20 20-25 25-30 30-35
Ave
rage
Lat
ency
of D
eliv
ered
Mes
sage
s
Message Event Interval
Message Event Interval vs Average Latency of DeliveredMessages
INFOCOM
RWP_98
6. Performance Evaluation
76
In Figure 6.8 (b) it is seen that AR delivery ratio is the lowest with RWP_98 due to
large number of connections compared to INFOCOM’06 and short time window size,
there is marginal change in temporal gain value. This switching between Forwarding
and Replication results into poor delivery of messages.
(a) INFOCOM’06 (b) RWP_98
Figure 6.8 Delivery Ratio Comparison of AR with Spray & Wait and PRoPHeT usingINFOCOM’06 and RWP_98
6.2.9 Overhead Ratio Comparison of AR with Spray & Wait andPRoPHeT using INFOCOM’06 and RWP_98
For Spray & Wait, overhead ratio is the lowest due to Direct Transmission in wait
phase, and the case of AR, switching contributes little to overhead. Here, switching is
less because of large time window size and smaller number of timestamps.
INFOCOM’06 has large number of connections, thus Encounter Based Forwarding
and Two Period Spray & Wait add additional overhead as shown in Figure 6.10(a).
AR overhead ratio is seen highest in Figure 6.10(b) because of constant switching
between Forwarding and Replication to adapt the network conditions, and it is due to
shorter time window size and large number of timestamps. Thus, more messages are
relayed and few of them are delivered. This is observed lowest in Spray & Wait as it
performs direct transmission in wait phase that adds minimum to overhead.
0
0.05
0.1
0.15
0.2
0.25
5-10 10-15 15-20 20-25 25-30 30-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
AR
PRoPHeT
SnW
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
5-10 10-15 15-20 20-25 25-30 30-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
AR
PRoPHeT
SnW
6. Performance Evaluation
77
(a) INFOCOM’06 (b) RWP_98
Figure 6.9 Overhead Ratio Comparison of AR with Spray & Wait andPRoPHeT using INFOCOM’06 and RWP_98
6.2.10 Comparing Number of Dropped Messages in AR with Spay& Wait and PRoPHeT using INFOCOM’06 and RWP_98
(a) INFOCOM’06 (b) RWP_98
Figure 6.10 Comparing Number of Dropped Messages in AR with Spay & Waitand PRoPHeT using INFOCOM’06 and RWP_98
Number of message drops in AR is less compared to PRoPHeT. INFOCOM’06 has
large number of connections and less number of timestamps. Thus, AR switches
almost equal number of times between Forwarding and Replication. Due to lower EV
0
100
200
300
400
500
600
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
AR
PRoPHeT
SnW
0
20
40
60
80
100
120
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
AR
PRoPHeT
SnW
0
200000
400000
600000
800000
1000000
1200000
5-10 10-15 15-20 20-25 25-30 30-35
Num
ber o
f Dro
pped
Mes
sage
s
Message Event Interval
Message Event Interval vs Number of DroppedMessages
AR
PRoPHeT
SnW
0
50000
100000
150000
200000
250000
300000
350000
5-10 10-15 15-20 20-25 25-30 30-35
Num
ber o
f Dro
pped
Mes
sage
s
Message Event Interval
Message Event Interval vs Number of DroppedMessages
AR
PRoPHeT
SnW
6. Performance Evaluation
78
values during Encounter Based Forwarding and multiple spraying opportunities in
Spray & Wait, though messages is relayed but gets dropped either due to TTL expiry
or limited buffer size. Thus, message drops in AR are higher compared to Spray &
Wait as shown 6.10(a).
AR overhead ratio is seen highest in Figure 6.10(b) due to constant switching between
Forwarding and Replication to adapt the network conditions. RWP_98 has shorter
time window size and large number of timestamps. This results into relay more
messages and few of them are delivered. Thus, number of messages drops is
observed for in AR with RWP_98.
6.2.11 Comparing Average Latency of Delivered Messages in ARwith Spray & Wait and PRoPHeT using INFOCOM’06and RWP_98
(a) INFOCOM’06 (b) RWP_98
Figure 6.11 Comparing Average Latency of Delivered Messages in AR withSpray & Wait and PRoPHeT using INFOCOM’06 and RWP_98
In Figure 6.11(a) the latency of delivered messages for AR is the lowest. This is
because INFOCOM’06 has a smaller number of timestamps and larger time window
size. AR gets enough opportunities through Encounter Based Forwarding and Two
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
5-10 10-15 15-20 20-25 25-30 30-35
Ave
rage
Lat
ency
of D
eliv
ered
Mes
sage
s
Message Event Interval
. Message Event Interval vs Average Latency of DeliveredMessages
AR
PRoPHeT
SnW0
1000
2000
3000
4000
5000
6000
5-10 10-15 15-20 20-25 25-30 30-35
Ave
rage
Lat
ency
of D
eliv
ered
Mes
sage
s
Message Event Interval
Message Event Interval vs Average Latency ofDelivered Messages
AR
PRoPHeT
SnW
6. Performance Evaluation
79
Period Spray & Wait for message delivery. Also, AR with INFOCOM’06 executes
Forwarding and Replication almost same number of times.
It is seen that AR latency of delivered message is the highest with RWP_98 in Figure
6.11(b) due to large number of connections and short time window size, and there is a
marginal change in temporal gain value. Thus, the switching between Forwarding and
Replication results into more overhead and higher delivery latency.
6.2.12 Comparing Delivery Ratio in AR with Spray & Wait andPRoPHeT using RollerNet and RWP_63
(a) RollerNet (b) RWP_63
Figure 6.12 Comparing Delivery Ratio in AR with Spray & Wait and PRoPHeTusing RollerNet and RWP_63
AR delivery ratio is the highest as shown in Figure 6.12(a) and 6.12(b), it is because,
the higher value of average temporal distance gain results into dense network
conditions. Thus, AR is able to execute Encounter Based Routing with large number
of EVs. Therefore, with the limited number of message copies in Forwarding phase,
AR is capable of delivering maximum messages for RollerNet and RWP_63.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
5-10 10-15 15-20 20-25 25-30 30-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
AR
PRoPHeT
SnW
0
0.02
0.04
0.06
0.08
0.1
0.12
5-10 10-15 15-20 20-25 25-30 30-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
AR
PRoPHeT
SnW
6. Performance Evaluation
80
6.2.13 Comparing Overhead Ratio in AR with Spray & Wait andPRoPHeT using RollerNet and RWP_63
AR’s overhead is marginally higher compared to Spray & Wait as shown in Figure
6.13(a), reason, the network is dense and AR performs Encounter Based Forwarding.
In this, the node compares its EV values with neighbor nodes before relaying
messages, and updates CWC at regular intervals, and adds to overhead.
In case of RWP_63 as shown in Figure 6.13(b) the overhead is seen highest. This is
due to less number of connections, AR switches back and forth, and is able to manage
good delivery ratio but with an additional overhead due to small number of
timestamps with large time window size compared to RollerNet.
(a) RollerNet (b) RWP_63
Figure 6.13 Comparing Overhead Ratio in AR with Spray & Wait and PRoPHeTusing RollerNet and RWP_63
6.2.14 Comparing Number of Dropped Messages in AR with Spray& Wait and PRoPHeT using RollerNet and RWP_63
Number of message drops is higher in AR compared to Spray & Wait as shown in
Figure 6.14(a). It is because the network is dense, but relayed messages are dropped
0
50
100
150
200
250
300
350
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
AR
PRoPHeT
SnW
0
10
20
30
40
50
60
70
80
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
AR
PRoPHeT
SnW
6. Performance Evaluation
81
due to TTL expiry or limited buffer size. In case of RWP_63, as shown in Figure
6.14(b) the overhead is seen highest which is due to less number of connections as AR
switches frequently. In Replication, it generates and relays message copies. Moreover,
messages are dropped due to TTL expiry or limited buffer size before it actually gets
delivered.
(a) RollerNet (b) RWP_63
Figure 6.14 Comparing Number of Dropped Messages in AR with Spray & Wait andPRoPHeT using RollerNet and RWP_63
6.2.15 Comparing Average Latency of Delivered Message in ARwith Spray & Wait and PRoPHeT using RollerNet andRWP_63
Latency of delivered messages for AR is little higher compared to PRoPHeT as
shown in Figure 6.15(a), because the network is dense and AR calls for Encounter
Based Routing. In this, the node compares the EV values before relaying the message
and updates CWC at regular intervals, and with medium latency message gets
delivered.
Figure 6.15(b) reflects that in RWP_63 the delivered message latency is seen higher
than the PRoPHeT. This is due to less number of connections as AR switches between
Forwarding and Replication. In Replication, it generates and relays message copies,
0
2000
4000
6000
8000
10000
12000
14000
16000
5-10 10-15 15-20 20-25 25-30 30-35
Num
ber o
f D
ropp
ed M
essa
ges
Message Event Interval
Message Event Interval vs Number of DroppedMessages
AR
PRoPHeT
SnW
0
200
400
600
800
1000
1200
5-10 10-15 15-20 20-25 25-30 30-35
Num
ber o
f Dro
pped
Mes
sage
s
Message Event Interval
Message Event Interval vs Number of DroppedMessages
AR
PRoPHeT
SnW
6. Performance Evaluation
82
but message drops due to TTL expiry or limited buffer size before it actually gets
delivered.
(a) RollerNet (b) RWP_63
Figure 6.15 Comparing Average Latency of Delivered Message in AR with Spray& Wait and PRoPHeT using RollerNet and RWP_63
Group-3: Comparison of AR using temporal closeness centrality with Spray& Wait and PRoPHeT
6.2.16 Comparing Delivery Ratio in AR with Spray & Wait andPRoPHeT under RollerNet and RWP_63 using TemporalCloness Centrality
AR delivery ratio is the highest as shown in Figure 6.16(a) and 6.16(b), because,
average closeness centrality value per time window is the higher for RollerNet
compared to RWP_63 resulting into dense network conditions. Thus, AR is able to
execute Encounter Based Routing with large number of EVs. Therefore, with the
limited number of message copies in Forwarding phase, AR is capable of delivering
maximum messages for RollerNet . RWP_63 due to random movements of nodes
overall delivery ratio is seen poor.
0
50
100
150
200
250
300
350
400
450
5-10 10-15 15-20 20-25 25-30 30-35
Ave
rage
Lat
ency
of D
eliv
ered
Mes
sage
s
Message Event Interval
Message Event Interval vs Average Latency ofDelivered Messages
AR
PRoPHeT
SnW0
200
400
600
800
1000
1200
1400
1600
5-10 10-15 15-20 20-25 25-30 30-35
Ave
rage
Lat
ency
of D
eliv
ered
Mes
sage
s
Message Event Interval
Message Event Interval vs Average Latency of DeliveredMessages
AR
PRoPHeT
SnW
6. Performance Evaluation
83
(a) RollerNet (b) RWP_63
Figure 6.16 Comparing Delivery Ratio in AR with Spray & Wait and PRoPHeTunder RollerNet and RWP_63 using Temporal Closeness Centrality
6.2.17 Comparing Overhead Ratio in AR with Spray & Wait andPRoPHeT under RollerNet and RWP_63 using TemporalCloness Centrality
AR’s overhead is marginally higher compared to PRoPHeT as shown in Figure
6.17(a), reason, the network is dense and AR performs Encounter Based Forwarding.
In this, the node compares its EV values with neighbor nodes before relaying
messages, and updates CWC at regular intervals, and adds to overhead.
In case of RWP_63 as shown in Figure 6.17(b) the overhead is seen highest. This is
due to less number of connections, AR switches back and forth with an additional
overhead
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
5-10 10-15 15-20 20-25 25-30 30-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
AR_CC
PRoPHeT
SnW
0
0.02
0.04
0.06
0.08
0.1
0.12
5-10 10-1515-2020-2525-3030-35
Del
iver
y R
atio
Message Event Interval
Message Event Interval vs Delivery Ratio
AR_CC
PRoPHeT
SnW
6. Performance Evaluation
84
(a) RollerNet (b) RWP_63
Figure 6.17 Comparing Overhead Ratio in AR with Spray & Wait and PRoPHeTunder RollerNet and RWP_63 using Temporal Closeness Centrality
6.2.18 Comparing Number of Dropped Messages in AR with Spray& Wait and PRoPHeT under RollerNet and RWP_63 usingTemporal Closeness Centrality
(a) RollerNet (b) RWP_63
Figure 6.18 Comparing Number of Dropped Messages in AR with Spray & Wait andPRoPHeT under RollerNet and RWP_63 using Temporal Closeness Centrality
0
50
100
150
200
250
300
350
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
AR
PRoPHeT
SnW0
10
20
30
40
50
60
70
80
5-10 10-15 15-20 20-25 25-30 30-35
Ove
rhea
d R
atio
Message Event Interval
Message Event Interval vs Overhead Ratio
AR
PRoPHeT
SnW
0
2000
4000
6000
8000
10000
12000
14000
16000
5-10 10-1515-2020-2525-3030-35
Num
ber o
f D
ropp
ed M
essa
ges
Message Event Interval
Message Event Interval vs Number of DroppedMessages
AR
PRoPHeT
SnW
0
100
200
300
400
500
600
700
800
900
Num
ber o
f Dro
pped
Mes
sage
s
Message Event Interval
Message Event Interval vs Number ofDropped Messages
AR_CC
PRoPHeT
SnW
6. Performance Evaluation
85
Number of message drops is higher in AR compared to Spray & Wait as shown in
Figure 6.18(a). It is because the network is dense, but relayed messages are dropped
due to TTL expiry or limited buffer size. In case of RWP_63, as shown in Figure
6.18(b) the overhead is seen highest which is due to less number of connections as AR
switches frequently. In Replication, it generates and relays message copies. Moreover,
messages are dropped due to TTL expiry or limited buffer size before it actually gets
delivered.
6.2.19 Comparing Average Latency of Delivered Messages in ARwith Spray & Wait and PRoPHeT under RollerNet andRWP_63 using Temporal Closeness Centrality
(a) RollerNet (b) RWP_63
Figure 6.19 Comparing Average Latency of Delivered Message in AR with Spray& Wait and PRoPHeT under RollerNet and RWP_63using Temporal Closeness
Centrality
Latency of delivered messages for AR is little higher compared to Spray & Wait as
shown in Figure 6.19(a), because the network is dense and AR calls for Encounter
Based Routing. In this, the node compares the EV values before relaying the message
0
50
100
150
200
250
300
350
400
450
5-10 10-15 15-20 20-25 25-30 30-35
Ave
rage
Lat
ency
of D
eliv
ered
Mes
sage
s
Message Event Interval
Message Event Interval vs Average Latencyof Delivered Messages
AR_CC
PRoPHeT
SnW0
200
400
600
800
1000
1200
1400
1600
5-10 10-15 15-20 20-25 25-30 30-35
Aver
age
Late
ncy
of D
eliv
ered
Mes
sage
s
Message Event Interval
Message Event Interval vs Average Latency ofDelivered Messages
AR_CC
PRoPHeT
SnW
6. Performance Evaluation
86
and updates CWC at regular intervals, and with medium latency message gets
delivered.
Figure 6.19(b) reflects that in RWP_63 the delivered message latency is seen higher
than the Spray & Wait. This is due to less number of connections as AR switches
between Forwarding and Replication. In Replication, it generates and relays message
copies, but message drops due to TTL expiry or limited buffer size before it actually
gets delivered.
Chapter Summary
Classification of IC-MANET routing is based on knowledge of future contacts and
number of message carriers i.e., Forwarding and Replication. It is seen that DASW
adapts the network conditions by setting up average delay and node degree. But, it is
static and limited to RollerNet. During our research it is established that the traces are
containing time varying information which was earlier underexplored. We have
evaluated RollerNet traces using Temporal Algorithm and accordingly DASW is
modified. Its performance is compared with Epidemic Routing for INFOCOM’05. It
is seen that there is little improvement in delivery ratio and lower overhead. Further,
Adaptive Routing is proposed that utilizes the temporal information for understanding
the network conditions. Basically, AR checks network density and decides to perform
either Encounter Based Forwarding or Two Period Spray and Wait. This is because,
when IC-MANET is dense Forwarding scheme works well and When Sparse,
Replication ensures limited copies are available for improving delivery ratio. Thus,
AR switches between Forwarding and Replication to adapt the network conditions
which improves routing efficiency.Temporal Algorithm is integrated with AR using
temporal distance and tempral closeness centrality. Based on these AR performance
analysis is carried out for real and synthetic datasets. Here, the objective was to
measure delivery ratio, overhead ratio, and number of dropped messages and average
latency of delivered messages. It is seen that the AR delivery ratio is better compared
to Spray & Wait and PRoPHeT using RollerNet traces, considerable for
INFOCOM’06. This confirms our findings that the evaluation of temporal properties
play a key role for efficient transmission and reception of message in IC-MANET.
Chapter 7
Summarization andConclusion
CHAPTER 7. SUMMARIZATION AND
CONCLUSION
7. Summarization and Conclusion
87
7SUMMARIZATION AND CONCLUSION
7.1 Summarization
It is known that in IC-MANET end to end connectivity does not exist. Therefore,
node mobility creates an opportunity for exchange of information since failure or
faults are not anomalies but an inherent part of the network.
Node mobility plays a vital role in challenging environments and one cannot ignore
the movement patterns related to properties, such as time order, frequency, contact
duration, inter contact time etc. Considering these as dynamic in nature. connections
are first analyzed and understood by using time varying graphs .
To understand the network conditions we have proposed Adaptive Routing that
utilizes the temporal distance and temporal centrality. The AR checks network density
and decides to perform either Encounter Based Forwarding or Two Period Spray &
Wait. This is because, when IC-MANET is dense Forwarding scheme works well and
7. Summarization and Conclusion
88
when Sparse, Replication ensures limited copies are available for improving delivery
ratio. Thus, AR switches between Forwarding and Replication to adapt the network
conditions and improve routing efficiency.
In the foregone chapters Temporal Characterization Algorithm has been analyzed and
designed for evaluating temporal properties. Because such framework helps in
computing number of time frames, size of time windows, distance and closeness
centrality. These properties are very useful in studying the dynamics of network
which helps in decision makings.
In the research we have integrated Temporal Algorithm with AR. The performance
analysis of AR is being carried out for real and synthetic datasets. Our objective was
to measure routing efficiency in terms of delivery ratio, overhead ratio, number of
dropped messages and average latency of delivered messages. The study reveals that
the AR delivery ratio is better compared to Spray & Wait and PRoPHeT using
RollerNet and considerable for INFOCOM’06. It confirms our findings that the
evaluation of temporal properties play a key role for efficient transmission and
reception of message in IC-MANET.
7.2 Conclusion
The present research work reveals that in challenging environment the temporal
properties play a vital role for the efficient diffusion of information.(Findigs are
published WORLCOMP’13)
Our research findings suggest that the algorithm can be applied to synthetic as
well as any real trace data available in a certain language format and data formats
so as to execute the algorithm effectively.
7. Summarization and Conclusion
89
Because the temporal properties (i.e., temporal distance and temporal closeness
centrality can be extracted from the datasets such as RollerNet, INFOCOM’06 )
are utilized by an AR protocol for understanding the dynamics of the network in
taking Forwarding or Replication decisions. (Findings are published and
presented WORLDCOMP’13)
As a result of switching the techniques between Forwarding and Replication
delivery ratio of AR seen is better compared to Spray & Wait and PRoPHeT.
(Findigs are published in International Journal of Communication and
Network, 2013)
AR is evaluated with RWP, RollerNet and INFOCOM’06 traces measure
delivery ratio, overhead and dropped messages. While comparing the results, it
establishes that the delivery ratio of an AR is higher for RollerNet trace than
RWP; it is due to lower connection edges and contact period. Number of contact
edges is higher in RollerNet trace resulting into lower overhead and dropped
messages than RWP.(Findigs are published in International Journal of
Communication and Network, 2013)
The performance of Spray & Wait, PRoPHeT and AR is evaluated for RollerNet
and INFOCOM’06 traces. For, RollerNet trace, AR’s delivery ratio is better
because of its large number of connections. In case of INFOCOM’06 trace, it is
due to the longer contact duration and shorter timestamps resulting higher
delivery ratio for AR compared to Spray & Wait and PRoPHeT.
7.3 Proposed Future Work
Temporal properties (temporal distance, temporal closeness centrality,
temporal betweenness) can be further analyzed to model Markov process
7. Summarization and Conclusion
90
based probabilistic behavior which could result in certain improvement in this
area. The same can be extended for energy efficient routing.
There is a scope to investigate the Denial of Service attack on AR routing and
propose accordingly the mechanism for prevention.
References
References
91
REFERENCES[1] C. D. J¨org Ott, Dirk Kutscher, “SIGCOMM’06 Workshops,” in Integrating
DTN and MANET Routing, 2006.
[2] J.-Y. L. B. J. Widmer, “ACM SIGCOMM Workshop on Delay-Tolerant
Networking (WDTN’05),” in Network coding for efficient communication in
extreme networks, 2005, pp. 284–291.
[3] K. Fall, “A delay-tolerant network architecture for challenged internets,” in
Proceedings of the 2003 conference on Applications, technologies,
architectures, and protocols for computer communications - SIGCOMM ’03,
2003.
[4] S. Guo, M. H. Falaki, E. A. Oliver, S. Ur Rahman, A. Seth, M. A. Zaharia, and
S. Keshav, “Very low-cost internet access using KioskNet,” ACM SIGCOMM
Computer Communication Review, vol. 37, no. 5, p. 95, Oct. 2007.
[5] A. Pentland, R. Fletcher, and A. Hasson, “DakNet: rethinking connectivity in
developing nations,” Computer, vol. 37, no. 1, pp. 78–83, Jan. 2004.
[6] T. Small and Z. J. Haas, “The shared wireless infostation model,” in
Proceedings of the 4th ACM international symposium on Mobile ad hoc
networking & computing - MobiHoc ’03, 2003.
[7] B. D. Davison, “Store-and-Forward Performance in a DTN,” in 2006 IEEE
63rd Vehicular Technology Conference, 2006, vol. 1, pp. 187–191.
[8] H. S. Yogeshwar P. Kosta, “Evolution of Various Controlled Replication
Routing Schemes for Opportunistic Networks,” in Advanced Computing
Communications in Computer and Information Science volume III, 2011, vol.
133, p. pp 338–347.
[9] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Efficient Routing in
Intermittently Connected Mobile Networks: The Multiple-Copy Case,”
IEEE/ACM Transactions on Networking, vol. 16, no. 1, pp. 77–90, Feb. 2008.
[10] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Efficient Routing in
Intermittently Connected Mobile Networks: The Single-Copy Case,”
IEEE/ACM Transactions on Networking, vol. 16, no. 1, pp. 63–76, Feb. 2008.
[11] V. Kostakos, “Temporal graphs,” Physica A: Statistical Mechanics and its
Applications, vol. 388, no. 6, pp. 1007–1023, Mar. 2009.
References
92
[12] N. Santoro, W. Quattrociocchi, P. Flocchini, A. Casteigts, and F. Amblard,
“Time-Varying Graphs and Social Network Analysis: Temporal Indicators and
Metrics,” Feb. 2011.
[13] J. Tang, M. Musolesi, C. Mascolo, and V. Latora, “Temporal distance metrics
for social network analysis,” in Proceedings of the 2nd ACM workshop on
Online social networks - WOSN ’09, 2009.
[14] M. E. J. Newman, “A measure of betweenness centrality based on random
walks,” Social Networks, vol. 27, no. 1, pp. 39–54, 2005.
[15] A. Clauset and N. Eagle, “Persistence and periodicity in a dynamic proximity
network,” DIMACS Workshop on Computational Methods for Dynamic
Interaction Networks, Nov. 2012.
[16] J. Tang, M. Musolesi, C. Mascolo, and V. Latora, “Characterising temporal
distance and reachability in mobile and online social networks,” ACM
SIGCOMM Computer Communication Review, vol. 40, no. 1, Jan. 2010.
[17] A. Keränen, T. Kärkkäinen, and J. Ott, “Simulating Mobility and DTNs with
the ONE (Invited Paper),” Journal of Communications, vol. 5, no. 2, pp. 92–
105, Feb. 2010.
[18] A. Lindgren, A. Doria, and O. Schelén, “Probabilistic routing in intermittently
connected networks,” ACM SIGMOBILE Mobile Computing and
Communications Review, vol. 7, no. 3, Jul. 2003.
[19] K. A. Harras and K. C. Almeroth, “Transport Layer Issues in Delay Tolerant
Mobile Networks,” in IN IFIP NETWORKING, 2006, pp. 463–475.
[20] S. Jain, K. Fall, and R. Patra, “Routing in a delay tolerant network,” ACM
SIGCOMM Computer Communication Review, vol. 34, no. 4, Oct. 2004.
[21] H. S. Yogeshwar P. Kosta, “Routing Enhancement Specific to Mobile
Environment Using DTN,” International Journal of Computer Theory and
Engineering, vol. 3, no. 4, pp. 537–542.
[22] K. A. Harras, K. C. Almeroth, and E. M. Belding-Royer, “Delay tolerant
mobile networks (DTMNs): controlled flooding in sparse mobile networks,” in
Proceedings of the 4th IFIP-TC6 international conference on Networking
Technologies, Services, and Protocols; Performance of Computer and
Communication Networks; Mobile and Wireless Communication Systems,
2005, pp. 1180–1192.
References
93
[23] M. Musolesi and C. Mascolo, “CAR: Context-Aware Adaptive Routing for
Delay-Tolerant Mobile Networks,” IEEE Transactions on Mobile Computing,
vol. 8, no. 2, pp. 246–260, Feb. 2009.
[24] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine, “MaxProp: Routing for
Vehicle-Based Disruption-Tolerant Networks,” in Proceedings IEEE
INFOCOM 2006. 25TH IEEE International Conference on Computer
Communications, 2006, pp. 1–11.
[25] V. Erramilli, M. Crovella, A. Chaintreau, and C. Diot, “Delegation
forwarding,” in Proceedings of the 9th ACM international symposium on
Mobile ad hoc networking and computing, 2008, pp. 251–260.
[26] P. Mundur, M. Seligman, and G. Lee, “Epidemic routing with immunity in
Delay Tolerant Networks,” in MILCOM 2008 - 2008 IEEE Military
Communications Conference, 2008, pp. 1–7.
[27] Y. Liao, K. Tan, Z. Zhang, and L. Gao, “Estimation based erasure-coding
routing in delay tolerant networks,” in Proceeding of the 2006 international
conference on Communications and mobile computing - IWCMC ’06, 2006.
[28] E. Bulut, Z. Wang, and B. K. Szymanski, “Cost Efficient Erasure Coding
Based Routing in Delay Tolerant Networks,” in 2010 IEEE International
Conference on Communications, 2010, pp. 1–5.
[29] Amin Vahdat and David Becker, “Epidemic Routing for Partially Connected
Ad Hoc Networks,” Tech. Rep. CS-200006, 2000.
[30] A. Beaufour, M. Leopold, and P. Bonnet, “Smart-tag based data
dissemination,” in Proceedings of the 1st ACM international workshop on
Wireless sensor networks and applications, 2002, pp. 68–77.
[31] A. Demers, D. Greene, C. Hauser, W. Irish, J. Larson, S. Shenker, H. Sturgis,
D. Swinehart, and D. Terry, “Epidemic algorithms for replicated database
maintenance,” in Proceedings of the sixth annual ACM Symposium on
Principles of distributed computing, 1987, pp. 1–12.
[32] Z. Li, L. Sun, and E. C. Ifeachor, “WSN10-5: Adaptive Multi-Copy Routing
for Intermittently Connected Mobile Ad Hoc Networks,” in Global
Telecommunications Conference, 2006. GLOBECOM ’06. IEEE, 2006, pp.
1–6.
[33] E. P. C. C. Jones, L. Li, P. A. S. S. Ward, and J. K. Schmidtke, “Practical
Routing in Delay-Tolerant Networks,” in Proceeding of the 2005 ACM
References
94
SIGCOMM workshop on Delay-tolerant networking - WDTN ’05, 2005, vol.
6, no. 8, pp. 943–959.
[34] Y. P. K. Hemal Shah, “Exploiting Wireless Networks, through creation of
Opportunity Network – Wireless-MobileAdhoc-Network (W-MAN) Scheme,”
International Journal of Ad hoc, Sensor & Ubiquitous Computing, vol. Vol. 2,
no. 1, pp. 99–110, 2011.
[35] R. H. Frenkiel, B. R. Badrinath, J. Borres, and R. D. Yates, “The Infostations
challenge: balancing cost and ubiquity in delivering wireless data,” IEEE
Personal Communications, vol. 7, no. 2, pp. 66–71, Apr. 2000.
[36] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Spray and Focus:
Efficient Mobility-Assisted Routing for Heterogeneous and Correlated
Mobility,” in Pervasive Computing and Communications Workshops, 2007.
PerCom Workshops ’07. Fifth Annual IEEE International Conference on,
2007, pp. 79–85.
[37] Y. Wang, S. Jain, M. Martonosi, and K. Fall, “Erasure-coding based routing
for opportunistic networks,” in Proceedings of the 2005 ACM SIGCOMM
workshop on Delay-tolerant networking, 2005, pp. 229–236.
[38] S. Jain, M. Demmer, R. Patra, and K. Fall, “Using redundancy to cope with
failures in a delay tolerant network,” ACM SIGCOMM Computer
Communication Review, vol. 35, no. 4, Oct. 2005.
[39] Y. Liao, Z. Zhang, B. Ryu, and L. Gao, “Cooperative Robust Forwarding
Scheme in DTNs using Erasure Coding,” in Military Communications
Conference, 2007. MILCOM 2007. IEEE, 2007, pp. 1–7.
[40] L.-J. Chen, C.-H. Yu, T. Sun, Y.-C. Chen, and H. Chu, “A hybrid routing
approach for opportunistic networks,” in Proceedings of the 2006 SIGCOMM
workshop on Challenged networks, 2006, pp. 213–220.
[41] E. M. Daly and M. Haahr, “Social network analysis for routing in
disconnected delay-tolerant MANETs,” in Proceedings of the 8th ACM
international symposium on Mobile ad hoc networking and computing -
MobiHoc ’07, 2007.
[42] P. Hui, J. Crowcroft, and E. Yoneki, “BUBBLE Rap: Social-Based
Forwarding in Delay-Tolerant Networks,” IEEE Transactions on Mobile
Computing, vol. 10, no. 11, pp. 1576–1589, Nov. 2011.
References
95
[43] P. Costa, C. Mascolo, M. Musolesi, and G. Picco, “Socially-aware routing for
publish-subscribe in delay-tolerant mobile ad hoc networks,” IEEE Journal on
Selected Areas in Communications, vol. 26, no. 5, pp. 748–760, Jun. 2008.
[44] F. Li and J. Wu, “LocalCom: a community-based epidemic forwarding
scheme in disruption-tolerant networks,” in Proceedings of the 6th Annual
IEEE communications society conference on Sensor, Mesh and Ad Hoc
Communications and Networks, 2009, pp. 574–582.
[45] T. Zhou, R. R. Choudhury, and K. Chakrabarty, “Diverse Routing: Exploiting
Social Behavior for Routing in Delay-Tolerant Networks,” in Proceedings of
the 2009 International Conference on Computational Science and Engineering
- Volume 04, 2009, pp. 1115–1122.
[46] Q. Li, S. Zhu, and G. Cao, “Routing in Socially Selfish Delay Tolerant
Networks,” in 2010 Proceedings IEEE INFOCOM, 2010, pp. 1–9.
[47] H. S. Yogeshwar P. Kosta, “Exploring and Exploiting Opportunistic Network
Routing in a DTN Environment,” International Journal of Future Generation
Communication and Networking, vol. 4, no. 2, pp. 13–21.
[48] W. Zhao, M. Ammar, and E. Zegura, “A message ferrying approach for data
delivery in sparse mobile ad hoc networks,” in Proceedings of the 5th ACM
international symposium on Mobile ad hoc networking and computing -
MobiHoc ’04, 2004.
[49] W. Zhao, Y. Chen, M. Ammar, M. Corner, B. Levine, and E. Zegura,
“Capacity Enhancement using Throwboxes in DTNs,” in 2006 IEEE
International Conference on Mobile Ad Hoc and Sensor Sysetems, 2006, pp.
31–40.
[50] J. Leguay, T. Friedman, and V. Conan, “DTN routing in a mobility pattern
space,” in Proceeding of the 2005 ACM SIGCOMM workshop on Delay-
tolerant networking - WDTN ’05, 2005, pp. 276–283.
[51] N. Sarafijanovic-Djukic, M. Pidrkowski, and M. Grossglauser, “Island
Hopping: Efficient Mobility-Assisted Forwarding in Partitioned Networks,” in
2006 3rd Annual IEEE Communications Society on Sensor and Ad Hoc
Communications and Networks, 2006, pp. 226–235.
[52] B. Burns, O. Brock, and B. N. Levine, “MV routing and capacity building in
disruption tolerant networks,” in Proceedings IEEE 24th Annual Joint
References
96
Conference of the IEEE Computer and Communications Societies., 2005, vol.
1, pp. 398–408.
[53] J. Zhang, G. Luo, K. Qin, and H. Sun, “Encounter-based routing in delay
tolerant networks,” in 2011 International Conference on Computational
Problem-Solving (ICCP), 2011, pp. 338–341.
[54] E. Bulut, Z. Wang, and B. K. Szymanski, “Cost-Effective Multiperiod
Spraying for Routing in Delay-Tolerant Networks,” IEEE/ACM Transactions
on Networking, vol. 18, no. 5, pp. 1530–1543, Oct. 2010.
[55] A. L. Traud, P. J. Mucha, and M. A. Porter, “Social Structure of Facebook
Networks,” CoRR, vol. abs/1102.2, 2011.
[56] J. Bonneau, J. Anderson, R. Anderson, and F. Stajano, “Eight friends are
enough: social graph approximation via public listings,” in Proceedings of the
Second ACM EuroSys Workshop on Social Network Systems, 2009, pp. 13–
18.
[57] C. Wilson, B. Boe, A. Sala, K. P. N. Puttaswamy, and B. Y. Zhao, “User
interactions in social networks and their implications,” in Proceedings of the
4th ACM European conference on Computer systems, 2009, pp. 205–218.
[58] H. Kwak, C. Lee, H. Park, and S. Moon, “What is Twitter, a social network or
a news media?,” in Proceedings of the 19th international conference on World
wide web, 2010, pp. 591–600.
[59] S. Scellato, C. Mascolo, M. Musolesi, and J. Crowcroft, “Track globally,
deliver locally: improving content delivery networks by tracking geographic
social cascades,” in Proceedings of the 20th international conference on World
wide web, 2011, pp. 457–466.
[60] E. Miluzzo, N. D. Lane, K. Fodor, R. Peterson, H. Lu, M. Musolesi, S. B.
Eisenman, X. Zheng, and A. T. Campbell, “Sensing meets mobile social
networks,” in Proceedings of the 6th ACM conference on Embedded network
sensor systems - SenSys ’08, 2008.
[61] E. Nordström, P. Gunningberg, and C. Rohner, “Haggle,” in Proceedings of
the 2009 MobiHoc S3 workshop on MobiHoc S3 - MobiHoc S3 ’09, 2009.
[62] N. Eagle, A. S. Pentland, and D. Lazer, “Inferring friendship network structure
by using mobile phone data.,” Proceedings of the National Academy of
Sciences of the United States of America, vol. 106, no. 36, pp. 15274–8, Sep.
2009.
References
97
[63] P.-U. Tournoux, J. Leguay, F. Benbadis, J. Whitbeck, V. Conan, and M. Dias
de Amorim, “Density-Aware Routing in Highly Dynamic DTNs: The
RollerNet Case,” IEEE Transactions on Mobile Computing, vol. 10, no. 12,
pp. 1755–1768, Dec. 2011.
[64] A.-L. Barabsi and R. Albert, “Emergence of Scaling in Random Networks,”
Science, vol. 286, no. 5439, pp. 509–512, 1999.
[65] D. Kotz, T. Henderson, I. Abyzov, and J. Yeo, “{CRAWDAD} data set
dartmouth/campus (v. 2009-09-09).” Sep-2009.
[66] M. Balazinska and P. Castro, “{CRAWDAD} data set ibm/watson (v. 2003-
02-19).” Feb-2003.
[67] M. Lenczner, B. Grgoire, and F. Proulx, “{CRAWDAD} data set
ilesansfil/wifidog (v. 2007-08-27).” Aug-2007.
[68] H. Ebel, L.-I. Mielsch, and S. Bornholdt, “Scale-free topology of e-mail
networks,” Physical Review E, vol. 66, no. 3, p. 035103, Sep. 2002.
[69] J. Shetty and J. Adibi, “Discovering important nodes through graph entropy
the case of Enron email database,” in Proceedings of the 3rd international
workshop on Link discovery, 2005, pp. 74–81.
[70] J. Leskovec and E. Horvitz, “Planetary-scale views on a large instant-
messaging network,” in Proceedings of the 17th international conference on
World Wide Web, 2008, pp. 915–924.
[71] D. J. Watts and S. H. Strogatz, “Collective dynamics of /`small-world/’
networks,” Nature, vol. 393, no. 6684, pp. 440–442, Jun. 1998.
[72] S. Wasserman and K. Faust, Social Network Analysis. Methods and
Applications. Cambridge University Press, 1994.
[73] L. Yang, H. Jiang, and J. Yang, “The Cyclic Temporal Network Density and
Impact on Information Diffusion for Delay Tolerant Network,” in 2010
International Conference on Computational Intelligence and Software
Engineering, 2010, pp. 1–5.
[74] J. Tang, M. Musolesi, C. Mascolo, V. Latora, and V. Nicosia, “Analysing
information flows and key mediators through temporal centrality metrics,” in
Proceedings of the 3rd Workshop on Social Network Systems - SNS ’10,
2010, pp. 1–6.
[75] A. Casteigts, P. Flocchini, W. Quattrociocchi, and N. Santoro, “Time-Varying
Graphs and Dynamic Networks,” CoRR, Nov. 2010.
Appendix – A
APPENDIX – A
Appendix - A
98
APPENDIX – A
Excerpts of communication exchanged with Dr. K.H. Lee during the literature studies.
from: Hemal Shah [email protected]
sender-time:
Sent at 9:15 PM (GMT+05:30). Current time there:11:57 AM.✆
date: Thu, Jun 17, 2010 at 9:15 PM
subject: Your kind assistance
mailed-by:
ganpatuniversity.ac.in
Dear Sir,
During these days, I worked two fold :
1. Understanding the mobility models and their simulation .
Outcome : I have updated myself about different mobility models and its operation /
behavior.
Query :1. On this URL
http://crawdad.cs.dartmouth.edu/meta.php?name=ncsu/mobilitymodels , there are
various traces (dataset) available. I found real traces interesting one. but,
a.) how do i analyze them ?
b.) What property shall i look into ?
c.) How do i analyze test this datasets
I tried with Mobility simulator called (mobisim) to generate traces [synthetic] and
evaluate it.but only small set of evaluation criteria availabe. ( like temporal chac,
spatial chac. relative speed ).
am interested for example to look into : Encounter time, Hitting time, Meeting time,
etc how to analyze for large dataset ?
Appendix - A
99
I think still i am investigetting for research gap in mobility model and its
characteristic. I read your publication on Levvyk walk and Slaw. How do i analyze it
? Is there any general framework available for analysis ?
Q.2 For DTN routing protocol, i am focusing on Spray and Wait.I tried to investigate
some issues with multicopy and have some idea like can we make it density aware ?
I used ONE simulator to re-simulate the result. Then i tried to test with Lee-slaw
dataset. But there is no provision to input external trace files to ONE. Infact, i put this
question to user group and still awaiting for answer .
Question: Is that possible to analyze the Slaw with Spray and Wait. i wish to figure
out some thing with node density.
With regards,
Hemal Shah (Ph.D.Scholar,GNU)
===================================================Reply
KHLee [email protected]
sender-time: Sent at 11:10 PM (GMT-04:00).Current time there: 2:28 AM.✆
to: Hemal Shah<[email protected]>
date: Thu, Jun 17, 2010 at 11:10 PM
subject: RE: Your kind assistance
mailed-by: gmail.com
signed-by: gmail.com
Dear Shah,
How are you doing? I’m struggling to find a critical factor besides the distance that
impacts the movement decision of humans. It’s too tough to figure out because of the
scarcity of raw data.
Appendix - A
100
For your inquiries,
1.Unfortunately, generating an analysis framework itself is an important research. So,
there is no specific way of doing that. It’s your freedom. Just remember that providing
a single quantification metric which differentiates previous models and real traces (i.e.
defining a gap between them) can be a great contribution in this field. I recommend
some metric related to temporal characteristic such as temporal distance (you
may find 1~2 papers in this issue at 2009~2010). The temporal characteristic is
clearly under explored and getting more attention even from physics guys.
2.I don’t think the research method testing a protocol over various mobility models is
meaningful. See my publication Max Contribution (infocom 2010) which tries to
design a scheduling protocol for Dense DTNs (where radio resource is not abundant
compared to the traffic volume). I used some knowledge in designing a protocol for
DTNs. I recommend you to develop a unique metric first and apply it to your
own protocol. Please, don’t spend time in simulations lacking your own ideas. Just
for example, combining with social network can be a promising area for
upcoming several years.
In a broader view, please keep in mind that top-down approach is much better than
bottom-up approach in recent network research trends. Imagine a bright future
exploiting novel networking technologies first and figure out what we really need in
our lives, then think about how we can realize the future. If you are not able to
imagine an attractive scenario first, you may suffer from writing an attractive
introduction of a paper. Tweaking a previous method is hardly accepted recently in
top conferences.
If you have an eye-opening idea in your aspect, feel free to discuss with me. I’ll be
glad to help you build up your scenarios.
I hope this helps
Best regards,
Kyunghan Lee
Appendix – BAPPENDIX – B
Appendix - B
101
APPENDIX – BExperimental Dataset Details
INFOCOM’05, INFOCOM’06 RollerNet real traces and RWP datasets are used for
temporal properties and routing protocol evaluation. Table B.1 summarized the
information about its location, duration, participants and devices used while collecting
the real traces.
Table B-1 Summary of Real and Synthetic Dataset Collections
1. Name INFOCOM’05 - cambridge/haggle/imote/intel (v. 2006-01-31)Location Intel Research Cambridge Corporate LaboratoryDate January 2005
DurationDevices distributed on Thursday, January 6, at 11:30amDevices collected on Tuesday, January 11, in the afternoon (mostof the traces last only for three days)
Participants16 admin staff, researchers, interns, and admin staff.iMote was left in the kitchen, as a stationary node, during theExperiment
Address IDsID 1 is the stationary node.ID 2-9 are corresponding to mobile iMotesID 10-128 corresponds to external devices
2. Name cambridge/haggle/imote/cambridge (v. 2006-01-31)Location Computer Lab, University of CambridgeDate End of January 2005
DurationDevices distributed on Tuesday, January 25th, 2005 at 14:00amDevices collected on Monday, January 31st, 2005 in the afternoon(most of the iMotes last around 5days)
Participants 19 graduate students from the System Research Group
Address IDsID 1-12 are corresponding to iMotes (Class 1)ID 13-223 corresponds to external devices (Class 2)
3. Name cambridge/haggle/imote/INFOCOM (v. 2006-01-31)Location Conference IEEE INFOCOM in Grand Hyatt MiamiDate March 2005
DurationDevices distributed on March 7th, 2005 between lunch time and5pm.Devices collected on March 10th, 2005 in the afternoon.
Participants 50 students, attending the student workshop
Address IDsID 1-41 are corresponding to iMotes (Class 1)ID 42-274 corresponds to external devices (Class 2)
4. Name cambridge/haggle/imote/INFOCOM’06 (v. 2009-05-29)Location Princesa Sofia Gran Hotel, Barcelona
Appendix - B
102
Date Monday, April 24th to Thursday April 27th, 2006
Duration
Devices distributed on Sunday April 23rd, between 7:00 and 9:00pm.Devices collected back starting from April 26th and on April 27thduring the day
Participants 70 students and researchers, attending the student workshop
Address IDs
ID 1-17 are static long range iMotesiMotes deployed throughoutthe area, ID 18-20 are long range iMotes that have been placed inlift of the hotel ID 21-98 are participants of the INFOCOMstudent workshop>ID100 are external devices
5. Name RollerNet-INFOCOM’09Location During roller tour in ParisDate August 20, 2006
DurationThe experiment started on Sunday, 20 Aug 2006 14:24:06(GMT),and stopped on Sunday, 20 Aug 2006 17:14:00 (GMT)
Participants2,500 people participated to the RollerbladingTour
Address IDs
ID 27-51 (25 imotes)Staff membersID 1-26 (26 iMotes)Skating associationsID 52-62 (11 iMotes)A set of friendsID 63-1112(1050 iMotes)external devices
6. Name RWP_98 (Generated using ONE Simulator)Duration 342915 SecondsParticipants 98 NodesMobilityModel
RWP
Interfacerange
100 meters
Address IDs n0 – n97
7. Name RWP_63(Generated using ONE Simulator)Duration 3096 SecondsParticipants 63 NodesMobilityModel
RWP
Interfacerange
100 meters
Address IDs n0 – n62
Appendix – C
APPENDIX – C
Appendix - C
103
APPENDIX – C
C.1 Customized Python script to convert real traces into ONE compatibleformat
Customized script converts synthetic dataset into connectivity report. First, input
parameters for RWP model set in ONE simulation as shown in Table B.1 serial
number 6 and 7.
<?php
include "connect.php";
$qry="select distinct source from detail order by source";
$result = mysql_query($qry,$cn);
//echo $result;
$i=0; $sim_end=43200;
while($data = mysql_fetch_array($result))
{ $nodes[$i]=$data[0]; $i++;}
$node_index=0;
$no_num=0;
while($node_index!=$i)
{ //FOR NODE is SOURCE
$no_num=0;
$qry="select * from detail where source='$nodes[$node_index]' order by desti";
$result = mysql_query($qry,$cn);
$nm=mysql_num_rows($result);
if($nm>0){
while($data=mysql_fetch_array($result)){
$no_num=1;
$ins_que="insert into arrange values('NULL',$data[1],'$data[2]','$data[3]','$data[4]','$data[5]')";
echo $ins_que;
mysql_query($ins_que,$cn);}}
if($no_num==1){
//DELETE ENTRY FOR MAIN DATABASE IF NODE ENTRY EXIT
$del_que="delete from detail where source='$nodes[$node_index]'";
mysql_query($del_que,$cn);}
//FOR NODE is DESTINATION
$no_num=0;
$qry="select * from detail where desti='$nodes[$node_index]' order by source";
$result = mysql_query($qry,$cn);
$nm=mysql_num_rows($result);
Appendix - C
104
if($nm>0{$no_num=1; while($data=mysql_fetch_array($result))
$ins_que="insert into arrange values('NULL',$data[1],'$data[2]','$data[4]','$data[3]','$data[5]')";
mysql_query($ins_que,$cn);}}
if($no_num==1){//DELETE ENTRY FOR MAIN DATABASE IF NODE ENTRY EXIT
$del_que="delete from detail where desti='$nodes[$node_index]'";
mysql_query($del_que,$cn);}
$node_index++;}$node_index=0;
$total_conn=0;$tcontact_time=0;
//ARRANGE THE DATASET (COMMON DATABASE FORMAT)
while($node_index!=$i){
$qry = "select * from arrange where source = '$nodes[$node_index]'";
$res = mysql_query($qry,$cn);
$nm = mysql_num_rows($res);$p=0;
if($nm>0){
$qry1="select distinct desti from arrange where source = '$nodes[$node_index]' order by desti";
echo "<br>".$qry1;
$result1=mysql_query($qry1,$cn);
$nm1=mysql_num_rows($result1);
if($nm1>0){while($data1=mysql_fetch_array($result1))
{$desti_list[$p]=$data1[0];$p++;}}}
$temp=0;
while($temp!=$p){$no_conn=0;$s_start=0;$s_end=0;
$qry1="select * from arrange where source = '$nodes[$node_index]' and desti =
$desti_list[$temp]' order by time";
$result1=mysql_query($qry1,$cn);
$nm1=mysql_num_rows($result1); $flag=0;$count=0;
if($nm1>0){ $action="up";while($data2=mysql_fetch_array($result1)){if($action=="up")
{$s_start=$data2[1];$action="down";$count++;$total_conn++;}
else{ $s_end=$data2[1];$action="up";$contact_time =$s_end -
$s_start;$tcontact_time =$tcontact_time;$contact_time; $ins_que="insert into trace_final
values('$nodes[$node_index]','$desti_list[$temp]',$s_start,$s_end,$count,$contact_time)";
mysql_query($ins_que,$cn);}}
if($action=="down"){$s_end=$sim_end;$contact_time =$s_end - $s_start;
$tcontact_time = $tcontact_time +
$contact_time;
$ins_que="insert into trace_final
values('$nodes[$node_index]','$desti_list[$temp]',$s_start,$s_end,$count,$contact_time)";
mysql_query($ins_que,$cn);
}}$temp++;} $node_index++;}
Appendix - C
105
//GENERARTE OUTPUT FILE THAT COMPTITABLE WITH TO EXTRACTING TEMPORAL
CHARACTERISTICS
$time_window_size = $tcontact_time/$total_conn;
$contents="Source,Destination,Start Time,End Time,No of Occurence,Interconnect Time,Windows
size, ".$time_window_size."\n";
$user_query = mysql_query("select * from trace_final",$cn);
while($row = mysql_fetch_array($user_query))
{$contents.=$row[0].",";
$contents.=$row[1].",";
$contents.=$row[2].",";
$contents.=$row[3].",";
$contents.=$row[4].",";
$contents.=$row[5].",";
$contents.="\n";
//$answer = str_replace(',', '\,', $row[faqdesk_answer_short]); // escape internalt commas
///$contents.=$answer."\n";}
$contents = strip_tags($contents); // remove html and php tags etc.
Header("Content-Disposition: attachment; filename=trace_final.csv");
print $contents;?>
Figure C.1 shows outcome of connections between nodes (node1, node2), i.e., Up or
down (action) and at what time (simulation time).
Figure C.1 ONE Connectivity Report
Appendix - C
106
C.2 Datasets Evaluation of Temporal Properties
Temporal characterization algorithm is implemented using networkX package in
python. Input to script dataset filename (. DAT file), Tmin (in seconds, Tmax (in
seconds), time window (in seconds) and output file (.txt) which contains evaluated
metrics from giving dataset (. DAT) file.
# Initialization of variable
i = 0;j = 0
k = 0;l = 0;x = 0
# Simulation Time
b = 3096;diameter = 0;hcent = 0
# Time Window Size
timewindow = 71
# Total Number of Nodes in Simulation
totalnode = 63
G = nx.Graph();G1 = nx.Graph();G2 = nx.Graph()
#Procedure for calculation of Distance in per Time stamps
def returndistance(source,target,stp):
temp = []
length = []
distance = 0
count1 = 0
for i in range(stp,(len(ar))):
if (source in ar[i]) and (target in ar[i]):
distance = 0 #print "Same Timestemp distance=",distance
return distance;elif (source in ar[i]) and (target not in ar[i]):
temp1 = (i+1)
[temp.append(x) for x in ar[i] if x not in temp] ;if i == (len(ar)-1):;
#print "on diff 1 Timestemp distance=",distance
distance = -1;return distance
for z in range((i+1),len(ar)):
length = [val for val in temp if val in ar[z]]
if (target not in ar[z]):
if (z == (len(ar)-1)):
distance = -1
#print "on diff 2 Timestemp distance=",distance
Appendix - C
107
return distance
else:[temp.append(x) for x in ar[z] if x not in temp]
elif ((target in ar[z]) and (len(length)==0)): if (z == len(ar)-1):
distance = 0
#print "on diff 3 Timestemp distance=",distance
return distance else:
[temp.append(x) for x in ar[z] if x not in temp]
elif ((target in ar[z]) and (len(length) > 0)) : distance = z
#print "on diff 4 Timestemp distance=",distance
return distance;break
elif ((i == (len(ar)-1)) and ((source not in ar[i]) or (target not in ar[i]))):
distance = 0
#print "on diff 5 Timestemp distance=",distance
return distance
else :
count1 = count1 + 1
if count1 >= len(ar):
distance = -1
#print "on diff 6 Timestemp distance=",distance
return distance
# Calculation of Highest Centrality per Time stamps
def highest_centrality(cent_dict):
if (len(cent_dict)>0):
cent_items=[(round(b,3),a) for (a,b) in cent_dict.iteritems()]
cent_items.sort()
cent_items.reverse()
hcent = tuple(reversed(cent_items[0]))
else:
hcent = 0
return hcent
Temporal Distance Calculation
def tempdist(source,target):
haha = []
stp = 0
rval = 0
rval = returndistance(source,target,stp)
if rval != -1:
haha.append(rval)
while (rval != -1 and rval != len(ar)):
rval = returndistance(source,target,rval)
Appendix - C
108
if rval != 0 and rval != -1:
haha.append(rval)
else:break
if (len(haha) > 0):
return (np.average(haha))
else:return -1
# Main Program Execution stating for Here
while(b < 3096):
tmin = 0
max = (timewindow + b)
tnotw = ((((tmax - tmin)/timewindow)))
ar = [[] for _ in range(tnotw)]
temporalmatrix = [[0 for row in range(0,totalnode)] for col in range(0,totalnode)]
wb = loadtxt('contacts.Exp2.dat') # datasets input file in common format
for rowx in wb:
i = int(rowx[2])
j = int(rowx[3])
k = int(rowx[0])
l = int(rowx[1])
m = int(rowx[4])
n = int(rowx[5])
if (1 <= l <= totalnode) and ((tmin <= i <= tmax) or ((i < tmin) and (tmin <= j <= tmax))):
G.add_edge(k,l)
ts = ((i-1)-tmin)/timewindow
if 1 <= l <= totalnode and 0 < ts+1 <= tnotw :
x = x + 1
if k not in ar[ts+1-1]:
[ts+1-1].append(k)
if l not in ar[ts+1-1]:
ar[ts+1-1].append(l)
for s in range(0,totalnode):
for t in range(0,totalnode):
ans = 0
ans = tempdist(s+1,t+1)
temporalmatrix[s][t] = ans
#print "TemporalDistanceMatrix =",temporalmatrix
summation = 0.0
nt = 0.0
for i in temporalmatrix:
for j in i:
Appendix - C
109
if j!= (-1):
summation = summation + j
cnt = cnt + 1
tlength = 0
c = 0
for g in nx.connected_component_subgraphs(G):
##print (nx.average_shortest_path_length(g))
tlength = tlength + (nx.average_shortest_path_length(g))
c = c + 1
## print "Total No. of connections occurs = ",x
## print "Total No. of edges in graph = ",G.number_of_edges()
print "Total no. of time window ",tnot
## G1 = (nx.dfs_tree(G))
## print "Avg Shortest Path Length",nx.average_shortest_path_length(G1)
## print "No. of Nodes in Graph = ",G.number_of_nodes()
## print "timestamp(in seconds) considered = ",timewindow
if(c>0):
staticdistancemetric = tlength/c
## print "Static average path length = ", staticdistancemetric
else:
staticdistancemetric = 0
## print "Static average path length = ", staticdistancemetric
averagetemporalmetric = summation/cnt
d = []
for g in nx.connected_component_subgraphs(G):
d.append(nx.diameter(g))
if (len(d) > 0):
diameter = max(d)
## print "Diameter of Graph = ",diameter
else:diameter = 0
## print "Diameter of Graph = ",diameter
# Compute degree centrality
dc = nx.degree_centrality(G)
## print sorted(dc.items(), key=itemgetter(1), reverse=True)
## print "highest degree centrality =",highest_centrality(dc)
# Compute betweenness centrality
bc = nx.betweenness_centrality(G)
## print sorted(bc.items(), key=itemgetter(1), reverse=True)
## print "highest betweenness centrality =",highest_centrality(bc)
# Compute closeness centrality
Appendix - C
110
cc = nx.closeness_centrality(G)
## print sorted(cc.items(), key=itemgetter(1), reverse=True)
## print "highest closeness centrality =",highest_centrality(cc)
####bc = nx.eigenvector_centrality(G)
####print sorted(bc.items(), key=itemgetter(1), reverse=True)
# store calculated parameter in output file Rollernet_3096_63.txt
f = open('c:/Python27/synthetic/Rollernet_3096_63.txt', 'a')
f.write('\n'+str(tmin)+'\t'+str(tmax)+'\t'+str(G.number_of_nodes())+'\t'+str(x)+'\t'+str(tnotw)+'\t'+str(tim
ewindow)+'\t'+str(round(staticdistancemetric,3))+'\t'+str(round(averagetemporalmetric,3))+'\t'+str(diam
eter)+'\t'+str(highest_centrality(dc))+'\t'+str(highest_centrality(bc))+'\t'+str(highest_centrality(cc)))
f.close()
b = (b + timewindow)
Using above script average temporal distance, diameter, degree centrality,
betweenness and Closeness centrality are evaluated for INFOCOM’05,
INFOCOM’06 and RWP_YY. This script is generalize to evaluate any real and
synthetic datasets.
C.3 Temporal Centrality Per Time Frame
Following script evaluates temporal centralities (Closeness and Betweeness ) per
time frame. We have used temporal closeness centrality for performance evaluation of
AR.
#include imporatant LiB
import networkx as nx
import matplotlib.pyplot as plt
from numpy import *
import numpy as np
from mmap import mmap,ACCESS_READ
from pylab import get_current_fig_manager, subplot, plot, legend, connect, show
from decimal import *
from operator import itemgetter
#INITILIZE VARIABLE
i = 0
j = 0
k = 0
Appendix - C
111
l = 0
x = 0
b = 0
diameter = 0
hcent = 0
timewindow = 132 #TIME FRAME SIZE
totalnode = 98 #TOTAL NUMBER OF NODES IN SIMULATION
#GRAPH VARIABLE
G = nx.Graph()
G1 = nx.Graph()
G2 = nx.Graph()
#TEMPORAL DISTANCE CALCULATION PROCEDURE
def returndistance(source,target,stp):
temp = []
length = []
distance = 0
count1 = 0
for i in range(stp,(len(ar))):
if (source in ar[i]) and (target in ar[i]):
distance = (i+1)
return distance
elif (source in ar[i]) and (target not in ar[i]):
temp1 = (i+1)
[temp.append(x) for x in ar[i] if x not in temp]
if i == (len(ar)-1):
distance = -1
return distance
for z in range((i+1),len(ar)):
length = [val for val in temp if val in ar[z]]
if (target not in ar[z]):
if (z == (len(ar)-1)):
distance = -1
return distance
else:
[temp.append(x) for x in ar[z] if x not in temp]
Elif ((target in air [z]) and (len(length)==0)):
if (z == len(ar)-1):
distance = 0
return distance
else:
Appendix - C
112
[temp.append(x) for x in ar[z] if x not in temp]
elif ((target in ar[z]) and (len(length) > 0)) :
distance = (z+1)
return distance
break
elif ((i == (len(ar)-1)) and ((source not in ar[i]) or (target not in ar[i]))):
distance = 0
return distance
else :
count1 = count1 + 1
if count1 >= len(ar):
distance = -1
#COMPUTE DEGREE CENTRALITY PER TIME WINDOW
def centrality_dc(cent_dict,t):
if (len(cent_dict)>0):
cent_items=[ b for (a,b) in cent_dict.iteritems()]
cent_items_node=[ a for (a,b) in cent_dict.iteritems()]
node_id=[ (a,b) for (a,b) in cent_dict.iteritems()]
if len(cent_items)== 0:
hcent = 0
else:
f1 = open('c:/Python27/synthetic/dc_centrality_r.txt', 'a')
f1.write('\n Time Window'+str(t))
for i in range(len(cent_items)):
p = cent_items_node[i]
p =int(float(p))
f1.write('\t Node'+str(p)+'='+str(cent_items[i]))
f1.close()
floatNums = [x for x in cent_items]
hcent = sum(floatNums) / len(cent_items)
else:
hcent = 0
return hcent
#COMPUTE BETWEENNESS CENTRALITY PER TIME WINDOW
def centrality_bc(cent_dict,t):
if (len(cent_dict)>0):
cent_items=[ b for (a,b) in cent_dict.iteritems()]
cent_items_node=[ a for (a,b) in cent_dict.iteritems()]
node_id=[ (a,b) for (a,b) in cent_dict.iteritems()]
if len(cent_items)== 0:
Appendix - C
113
hcent = 0; else:
f1 = open('c:/Python27/synthetic/bc_centrality_r.txt', 'a')
f1.write('\n Time Window'+str(t))
for i in range(len(cent_items)):
p = cent_items_node[i]
p =int(float(p))
f1.write('\t Node'+str(p)+'='+str(cent_items[i]))
f1.close()
floatNums = [x for x in cent_items]
hcent = sum(floatNums) / len(cent_items)
else:
hcent = 0
return hcent
#COMPUTE CLOSENESS CENTRALITY PER TIME WINDOW
def centrality_bc(cent_dict,t):
if (len(cent_dict)>0):
cent_items=[ b for (a,b) in cent_dict.iteritems()]
cent_items_node=[ a for (a,b) in cent_dict.iteritems()]
node_id=[ (a,b) for (a,b) in cent_dict.iteritems()]
if len(cent_items)== 0:
hcent = 0
else:
f1 = open('c:/Python27/synthetic/cc_centrality_r.txt', 'a')
f1.write('\n Time Window'+str(t))
for i in range(len(cent_items)):
p = cent_items_node[i]
p =int(float(p))
f1.write('\t Node'+str(p)+'='+str(cent_items[i]))
f1.close()
floatNums = [x for x in cent_items]
hcent = sum(floatNums) / len(cent_items)
else:
hcent = 0
return hcent
#PROCEDURE TO FIND THE HIGHEST CENTRALITY PER TIME FRAME
def highest_centrality(cent_dict):
if (len(cent_dict)>0):
cent_items=[(b,a) for (a,b) in cent_dict.iteritems()]
cent_items.sort()
cent_items.reverse()
Appendix - C
114
hcent = tuple(reversed(cent_items[0]))
else:
hcent = 0
return hcent
def tempdist(source,target):
haha = []
stp = 0
rval = 0
rval = returndistance(source,target,stp)
if rval != -1:
haha.append(rval)
while (rval != -1 and rval != len(ar)):
rval = returndistance(source,target,rval)
if rval != 0 and rval != -1:
haha.append(rval)
else:
break
if (len(haha) > 0):
return (np.average(haha))
else:
return -1
while(b < 342916):
tmin = 0
tmax = (timewindow + b)
tnotw = ((((tmax - tmin)/timewindow)))
ar = [[] for _ in range(tnotw)]
temporalmatrix = [[0 for row in range(0,totalnode)] for col in range(0,totalnode)]
wb = loadtxt('sythetic_63.dat') # INPUT TRACE FILE NAME
for rowx in wb:
i = int(rowx[2])
j = int(rowx[3])
k = int(rowx[0])
l = int(rowx[1])
m = int(rowx[4])
n = int(rowx[5])
if (1 <= l <= totalnode) and ((tmin <= i <= tmax) or ((i < tmin) and (tmin <= j <= tmax))):
G.add_edge(k,l)
ts = ((i-1)-tmin)/timewindow
if 1 <= l <= totalnode and 0 < ts+1 <= tnotw :
x = x + 1
Appendix - C
115
if k not in ar[ts+1-1]:
ar[ts+1-1].append(k)
if l not in ar[ts+1-1]:
ar[ts+1-1].append(l)
for s in range(0,totalnode):
for t in range(0,totalnode):
ans = 0
## print "started."
ans = tempdist(s+1,t+1)
## print "source =",s+1," destination =",t+1," ans =",ans
temporalmatrix[s][t] = ans
##print "TemporalDistanceMatrix =",temporalmatrix
summation = 0.0
cnt = 0.0
for i in temporalmatrix:
for j in i:
if j!= (-1):
summation = summation + j
cnt = cnt + 1
tlength = 0
c = 0
for g in nx.connected_component_subgraphs(G):
tlength = tlength + (nx.average_shortest_path_length(g))
c = c + 1
print "Total no. of time window ",tnotw
if(c>0):
staticdistancemetric = tlength/c
else:
staticdistancemetric = 0
averagetemporalmetric = summation/cnt
d = []
for g in nx.connected_component_subgraphs(G):
d.append(nx.diameter(g))
if (len(d) > 0):
diameter = max(d)
else:
diameter = 0
# Compute Degree centrality
dc = nx.degree_centrality(G)
# Compute Betweenness centrality
Appendix - C
116
bc = nx.betweenness_centrality(G)
# Compute closeness centrality
cc = nx.closeness_centrality(G)
#WRITE EXTRACTED CHARACTERISTIC IN OUTPUT FILE
f = open('c:/Python27/synthetic/Synthetic_63.txt', 'a')
f.write('\n'+str(tmin)+'\t'+str(tmax)+'\t'+str(G.number_of_nodes())+'\t'+str(x)+'\t'+str(tnotw)+'\t'+str(tim
ewindow)+'\t'+str(staticdistancemetric)+'\t'+str(averagetemporalmetric)+'\t'+str(diameter)+'\t'+str(highe
st_centrality(dc))+'\t'+str(highest_centrality(bc))+'\t'+str(highest_centrality(cc)))
f.close()
b = (b + timewindow)
Appendix – D
APPENDIX – D
Appendix - D
117
APPENDIX – D
D 1 AverageTemporal Distance values for RWP_63
Temporal Algorithm builds temporal distance matrix and performs the summation of
non-negative values of matrix. Using equation 4.2, it computes the average temporal
distance.Table D-1 shows the computed average temporal distance values for
RWP_63.
Table D-1 Average Temporal Distance Values for RWP_63
Timestamp Number
TemporalDistance
1 02 03 04 05 06 07 08 09 010 011 012 0.00613 0.01314 0.013
15 0.03216 0.03217 0.03218 0.0719 0.0720 0.08121 0.08122 0.12423 0.12424 0.12425 0.12426 0.12427 0.13528 0.13529 0.13530 0.148
31 0.16332 0.16333 0.16334 0.16335 0.16336 0.21537 0.22938 0.22939 0.22940 0.22941 0.22942 0.22943 0.31744 0.317
Average 0.105
D 2 Temporal Closeness Centrality values for experimental datasets
Temporal closeness centrality values are computed for INFOCOM’06, RollerNet and
RWP_63, in which the table column shows a number of time stamps, node id and its
centrality value per timestamp.
It is observed in D.1 (a) to (k) that the static value of centrality only considers node
ids with highest centrality values from entire simulation duration, wherein, temporal
Appendix - D
118
closeness centrality values shows the node id with centrality value per timestamps.
Thus, closeness centrality values may vary from current time stamp to next.
INFOCOM ‘06
Temporal Cloness Centrality values are calculated for INFOCOM’06 and presented
in Table D.2 (a) to (c).
Table D 2 (a) (b) (c)
1 0 0
2 1 1
3 4 0.63
4 5 0.636
5 4 0.642
6 4 0.648
7 4 0.65
8 4 0.654
9 4 0.659
10 3 0.663
11 3 0.664
12 3 0.664
13 3 0.665
14 3 0.665
15 3 0.665
16 3 0.662
17 3 0.759
18 3 0.789
19 3 0.805
20 3 0.806
21 4 0.783
22 3 0.797
23 3 0.808
24 3 0.817
25 2 0.82
26 2 0.826
27 2 0.828
28 2 0.834
29 2 0.836
30 2 0.838
31 2 0.838
32 2 0.839
33 2 0.839
34 2 0.84
35 2 0.84
36 2 0.84
37 2 0.84
38 2 0.841
39 2 0.841
40 2 0.841
41 2 0.841
42 2 0.841
43 2 0.842
44 2 0.855
45 2 0.867
46 2 0.872
47 2 0.884
48 2 0.887
49 2 0.898
50 2 0.902
51 2 0.907
52 2 0.911
53 2 0.914
54 2 0.916
55 2 0.917
56 2 0.917
57 2 0.918
58 2 0.918
59 2 0.918
60 2 0.918
61 2 0.918
62 2 0.918
63 2 0.918
64 2 0.918
65 2 0.918
66 2 0.918
67 2 0.918
68 2 0.918
69 2 0.918
70 2 0.92
71 2 0.922
72 2 0.924
73 2 0.926
74 2 0.928
75 2 0.93
76 2 0.931
77 2 0.932
78 2 0.934
79 2 0.934
80 2 0.935
81 2 0.936
82 2 0.936
83 2 0.936
84 2 0.936
85 2 0.936
86 2 0.936
87 2 0.936
88 2 0.936
89 3 0.838
Appendix - D
119
RollerNet
Temporal Cloness Centrality values are calculated for RollerNet and presented in
Table D.2 (d) and (e).
Table D 2 (d) (e)1 9 0.409
2 7 0.442
3 5 0.525
4 4 0.568
5 5 0.483
6 4 0.509
7 3 0.538
8 3 0.553
9 3 0.571
10 4 0.58
11 3 0.584
12 3 0.586
13 3 0.596
14 3 0.608
15 3 0.62
16 3 0.625
17 3 0.63
18 3 0.633
19 3 0.637
20 3 0.638
21 3 0.643
22 3 0.65
23 3 0.652
24 3 0.655
25 3 0.659
26 3 0.662
27 3 0.664
28 3 0.666
29 3 0.67
30 3 0.672
31 3 0.676
32 3 0.68
33 3 0.683
34 3 0.688
35 3 0.691
36 3 0.694
37 3 0.697
38 3 0.7
39 3 0.704
40 3 0.707
41 3 0.709
42 3 0.712
43 3 0.715
44 3 0.719
RWP_63
Temporal Cloness Centrality values are calculated for RWP_63 and presented in
Table D.2 (f) to (h).
Table D 2 (f) (g) (h)
1 1 1
2 1 1
3 1 1
4 1 1
5 1 1
6 1 1
7 1 1
8 1 1
9 1 1
10 1 1
11 1 1
12 3 0.833
13 3 0.807
14 3 0.807
15 3 0.753
16 3 0.753
17 5 0.626
18 5 0.532
19 4 0.578
20 4 0.556
21 4 0.556
22 4 0.549
23 4 0.549
24 4 0.549
25 4 0.542
26 4 0.542
27 6 0.4
28 6 0.397
29 6 0.397
30 6 0.398
31 6 0.289
32 6 0.289
33 6 0.289
Appendix - D
120
34 6 0.289
35 6 0.289
36 6 0.286
37 6 0.284
38 6 0.284
39 6 0.284
40 6 0.29
41 6 0.29
42 6 0.29
43 6 0.285
44 6 0.285
List of Publications
List of Publications
121
LIST OF PUBLICATIONS
International Journal
1. H. Shah, Y. Kosta and V. Patel, "Efficient Routing Using Temporal Distance
in Intermittently Connected Mobile Ad-hoc Networks," Communications
and Network, Vol. 5 No. 3, 2013, pp. 264-271. doi: 10.4236/cn.2013.53033.
2. Hemal Shah, Yogeshwar Kosta P. Article: Exploiting Wireless Networks,
through creation of Opportunity Network – Wireless-Mobile-Ad-hoc-
Network (W-MAN) Scheme", International Journal of Ad hoc, Sensor &
Ubiquitous Computing(IJASUC), March-2011, Vol:2 No.1, pp. 99- 110,
http://airccse.org/journal/ijasuc/current2011.html.
3. Hemal Shah, Yogeshwar Kosta P Article: "Exploring and Exploiting
Opportunistic Network Routing in a DTN Environment." International
Journal of Future Generation Communication Networks," Korea, Vol.4 No.-
2 July 2011, http://www.sersc.org/journals/IJFGCN/vol4_no2/2.pdf.
International Conference
1. H. Shah , Yogeshwar Kosta, “Characterization and Evaluation: Temporal
Properties of Real and Synthetic Dataset for DTN”, International
Conference of Wireless Network track, WorldComp’13, USA,2013 pp.206-
213
2. Hemal Shah, Yogeshwar Kosta, "Evolution of various controlled based
replication routing schemes for opportunistic networks,” AUC 2.0
International Conference, Published by Springer Computer and Information
Science (CCIS) Series, 2011, Part -III, pp. 337-359.
Communications and Network, 2013, 5, 264-271 doi:10.4236/cn.2013.53033 Published Online August 2013 (http://www.scirp.org/journal/cn)
Copyright © 2013 SciRes. CN
Efficient Routing Using Temporal Distance in Intermittently Connected Mobile Ad-hoc Networks
Hemal Shah, Yogeshwar Kosta, Vikrant Patel Faculty of Computer Engineering, Ganpat University, Mehsana, India
Email: [email protected], [email protected], [email protected]
Received June 19, 2013; revised July 19, 2013; accepted August 10, 2013
Copyright © 2013 Hemal Shah et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ABSTRACT The analysis of real social, biological and technological networks has attracted a lot of attention as technological ad- vances have given us a wealth of empirical data. For, analysis and investigation time varying graphs are used to under- stand the relationship, contact duration, repeated occurrence of contact. It is under exploring in intermittently connected networks. Now, by extending the same concept in intermittent networks, the efficiency of the routing protocol can be im- proved. This paper discusses about the temporal characterizing algorithm. Such characterization can help in accu-rately understanding dynamic behaviors and taking appropriate routing decisions. Therefore, the present research pro-vokes ex- ploring different possibilities of utilizing the same time varying network analyses and designing an Adaptive Routing protocol using temporal distance metric. The adaptive routing protocol is implemented using ONE simulator and is compared with the Epidemic and PropHET for delivery ratio, overhead and the number of dropped messages. The result reveals that Adaptive routing performs better than Epidemic and PropHET for real and synthetic datasets. Keywords: Temporal Graph; Temporal Distance; DTN; Forwarding; Replication
1. Introduction In today’s world, connectivity via the Internet is an inte- gral part to connect, share and communicate information across most devices and systems, whether mobile-device or otherwise. We know, determined connectivity is nei- ther a rule nor a mandatory practice and most wireless applications demand stringent operating or connectivity attributes. In today’s wireless environment, comprising of Micro and Pico networks, such as, vehicular networks, pocket-switched networks, etc.; dynamic or otherwise, operate in a hostile environment, where network para- meters are constantly changing in terms of time, rate, order and proportion of known-unknown and unknown- unknown kinds that lead to losses. This makes network behavioral-response difficult to predict, where, failure or faults are not anomalies but rather an integral part of a dynamic network systems [1]. In this paper, we success- fully articulate and report the modeling aspects that help to predict the behavioral-response of such networks, characterized using our model-approach. Our research exploits mobility associated with the nodes to establish connectivity and affect data transfer. Our proposed model of intermittently connected-mobile ad-hoc network (IC- MANET) [2] is realized using attributes such as time
varying graphs [3] and temporal distance [4]. Using these we have designed a novel temporal algorithm to charac- terize the response of the network. The designed algo- rithm was utilized to validate, INFOCOM’06 [5], Rol- lerNet [6]—real traces and Random Way Point [7] to establish the concept by calculating the number of time- frame, time window size, temporal distance etc. Parame- ters obtained were then utilized to improve the routing efficiency of IC-MANET. Adaptive Routing (AR) util- izes the temporal distance for accurate forwarding deci- sions and understanding the dynamics associated with such network by integrating encounter based forwarding and two periods of spray and wait based replication tech- niques. By applying our concept, which exploits the node mobility and lower overhead to networks; our findings reveal significant improvement in the packet delivery ratio.
Section 2 discusses about related theories of time va-rying graphs and defines temporal distance metric. Sec-tion 3 discusses the design of the temporal algorithm and its application real traces. Section 4 presents the design and development of AR using temporal distance, simula-tion and performance analyses. Section 5 presents con-clusions and future work.
H. SHAH ET AL.
Copyright © 2013 SciRes. CN
265
2. Temporal Graph In static or aggregated networks, it is observed [8] that the connections (in case of ad-hoc network/mobile ad- hoc network) are inherently varying over time and ex- hibit more dimensionality [4] than static analysis can capture. Static graphs treat all links as appearing at the same time. It is unable to capture key temporal charac- teristics, and gives an overestimate of potential paths, connection pairs of nodes which cannot provide any in- formation on the delay associated with an information spreading process. Thus, to represent IC-MANET as temporal graphs, the mobile nodes can be presented as vertices and opportunistic contact between nodes as an edge. In the understanding duration of contact, inter- contact time, repeated contact, the time order of contact along a path on time interval basis, Temporal graph [3] is represented by sequence of time windows, for each win- dow is considered a snapshot of the network at that time interval.
Consider the sequence of interaction. From this we can construct the example temporal graph (Figure 1(a)) and corresponding static aggregated graph (Figure 1(b)), where interactions between a pair of nodes defines an edge or, equivalently, generated from the union of all edges in the temporal graph [3]. Next, let’s define the definition of temporal graph, path and distance.
2.1. Temporal Graph Given a network trace starting at minT and ending at
maxT , a contact between nodes, i, j at time “s” is defined with the notation s
ijR . A temporal graph [3] ( )min max,t T Tω with N nodes consists of a sequence of
graphs mintG ,
minT wG + ,…, maxTG , where “w” is the size
of each time window unit e.g., seconds. Then, tG con- sists of a set of nodes V and a set of edges E such that ,i j V∈ , if and only if, there exists s
ijR with t s t w≤ ≤ + .
2.2. Temporal Path For given two nodes i and j temporal path defines as:
(a) (b)
Figure 1. Example of Temporal Graph with three timewin- dows and six nodes. (a) Temporal Graph; (b) Static Graph.
( )min max,Thijp T (1)
To be the set of paths starting from i and finishing at j that passes through the nodes 1
1itt
in n , where 1i it t− ≤ and min i maxT t T≤ ≤ is the time window, that node n is visited and h is the maximum hops within the same win-dow t. There may be more than one shortest path.
2.3. Temporal Distance Given two nodes i and j, the shortest temporal distance defines as:
( )min max,hijd T T (2)
To be the shortest temporal path length, starting from time minT , this can be thought as the number of time windows (or temporal hops) which takes for information to spread from a node i to node j. The horizon h indicates the maximum number of nodes within each window TG through which information can be exchanged, or in prac-tical terms, the speed that a message travels. In the case of temporally disconnected node pairs q, p i.e., informa-tion from q never reaches p, then set the temporal dis-tance pqd = ∞ .
3. Temporal Algorithm Temporal distance ( )min maxijd T ,T , is computed in terms of number of time windows i.e.
( ) ( )min max min max, ,tij ijd T T d T T= . For each pair of i and j,
algorithm computes ( )min max,ijd T T and then, takes av- erage of all values. This way temporal distance is com-puted in number of time stamps. If average value multip-lies with w, then result is the temporal distance in terms of time (in seconds). Equation (1) gives average temporal distance between minT and maxT :
( ) ( ) ( )min max min max, ,1 ij
ijL T T d T T
N Nω
=− ∑ (3)
3.1. Timewindow (w) Calculation To understand the computation of timewindow, refer Ta-ble 1 below showing calculation on dataset as an exam-ple, where each cell value represents the total contact time between a particular pair i, j divided by total the number of contact occurrences. For each node pair (i, j) compute a sum of all values. It returns the average meet-ing time per contact. The optimal value of time window is greater than average meeting time, because if time window ≤ average meeting time, then in most of the time windows, number of contact occurrence will be around one. That means the information cannot be diffused effi-ciently into the network.
H. SHAH ET AL.
Copyright © 2013 SciRes. CN
266
Table 1. Timewindow Calculation.
Node ID 1 2 ∑ Total Contact Time
Total No. of Occurances
1 0/0 480/2 480/2
2 500/2 0/0 500/2
( )min max,ij
ij
T T TN
∑∑
9804
= 245 Time Window size
From Table 1 computation, it is established that, for
effective information diffusion process into the network optimal timewindow should be greater than
( )min max,ij
ij
T T TN
∑∑
. E.g., in Figure 1(a) total number of
time ( ) ( )max minwindow 900 0 300 3T T w= − = − = timestamps, assuming timewindow size = 300. Let’s find temporal distance ( )0 900,t
ijd t t for the temporal graph shown in Figure 1(a). Here, minT = 0 and maxT = 900. Timewindow size = 300. Thus, there are three time windows 1t , 2t and 3t .
3.2. Computation of Temporal Distance Before starting calculation of temporal distance of each pair i, j, initialize number of empty lists equal to that of calculating number of time window. For each pair (i, j), i ≠ j, start scanning timestamps from 1 to 3. For each timestamp, add occurred node id into the respective list of timestamp. Pair of node (i, j) occurs whenever there is a contact edge between node pair (i, j).
Preconditions Pair of node (i, j) occurs whenever there is a contact
edge between node pair (i, j). Case 1: If i == j then, return 0, in computing matrix
below, temporal distance (A, A) = (B, B) = (C, C) = (D, D) = 0
Case 2: If both i and j occurs in same timestamp then return (jth timestamp number—ith timestamp number) or return (0). In Figure 1(a), node A and node B occurs in same timestamp no. 1, so the temporal distance between A and B is (B’s timestamp no. − A’s timestamp no.) = (1 − 1) = 0 timestamps.
Case 3: If i occurs earlier than j, then search occur-rences of j in consecutive timestamps by using other oc-curred nodes in same timestamp in which i has occurred; for each pair i, j it may give more than one path in terms of required timestamp, in such a case select the shortest timestamp. In Figure 1(a), for temporal distance (A, D), node A occurred in timestamp number 1 and node D oc-curred in timestamp number 3. Also, there is an interme-diate node B which is common between node A and node D. So temporal distance (A, D) = (node D’s timestamp number − node A’s timestamp number) = (3 – 1) = 2
timestamps. Case 4: If i occurs and j do not occur during a consec-
utive timestamp till maxT , then the temporal path be-tween a pair of i, j is not possible. So, return ∞. Figure 1(a), for temporal distance (D, E), node D occurred in timestamp number 3. But there are no occurrences of node E also by using other intermediate occurrences of other nodes. So temporal distance (D, E) = ∞. In contin-uation of the example shown in Figure 1(a) and succes-sive computation of temporal distance matrix as above, the sum of non-negative values of matrix = 10. Now, calculate the average temporal distance metric: 300 (10/(6)(5)) =3000/(30) =100. i.e., it takes average 100 seconds to reach from source “i” to destination “j”.
Temporal Distance Matrix0 0 1 1 1 10 0 1 1 1 11 1 0 1 0 01 0 0 0 1 11 1 0 1 0 01 1 0 1 0 0
− − − − −
= − − − − −
3.3. Temporal Algorithm
1) Input source and target, minT and maxT time window Time Window Equation:
( ) ( )min max,ij
ij
T T Tw Timewindow
N> ∑
∑
Where, ( )min max,ijT T T∑ = Total contact time be-
tween all pair of nodes i, j and ijN∑ = Total occur-rences of all pairs of nodes i and j.
2) Number of times frames = max minT T Time− win- dow.
3) Initialize number of empty list equal to number of time frames. Each list shows node ids whose contact oc-curred in a respective time frame.
4) Read the dataset and perform a lookup for node contact in different time frames and generate a distance matrix for each node. Per contact frame, fill up the array/ list with node ids in contact.
5) Compute the temporal distance as: a) If source and target ids are in the same list, return
(target time frame number—source time frame number) as temporal distance.
b) Otherwise, look up source and target in different time frames. If the source time frame < target time frame then return (target time frame number − source time frame number) as temporal distance.
c) In case repeated occurrence of the source, target sets minT = last target occurred + 1 timestamp and repeat steps
H. SHAH ET AL.
Copyright © 2013 SciRes. CN
267
(a) and (b). 6) Take average values of all pairs (source, target)
temporal distance. 7) Repeat steps 4, 5, 6 and 7 for all pairs (source, tar-
get) and generate matrix. Minus one (−1) indicates no edge between a pair of nodes in the matrix.
3.4. Application of Temporal Algorithm to INFOCOM’06, RollerNet—Real Trace and RWP Synthetic Datasets
First network topology is generated from large real data sets using python custom made script. Python provides a module called networkX, which helps to generate net-work topology according to the dataset. For evaluation, we have downloaded the INFOCOM’06, RollerNet real trace and RWP dataset from CRAWDAD. Tables 2, 3, 4 and 5 present duration, start and end time of trace, num-ber of node participated and total number of contacts occurred between nodes for experimental datasets. Note here RWP_XX denotes number of nodes taken of real and synthetic dataset. Throughout our discussion we have compared temporal metrics values of real and synthetic to differentiate the results and to highlight the importance of real traces.
Table 2. INFOCOM’06 Dataset details.
Duration Start-End Time Number of Nodes
Total Number of Contacts
4 days
Day 1: 61,260 - 86,400 (6.98 hours)
98 118,875 (of all four days)
Day 2: 86,400 - 172,800 (24 hours)
Day 3: 172,800 - 259,200 (24 hours)
Day 4: 259,200 - 345,600 (24 hours)
Table 3. RWP_98 Dataset details.
Duration Start-End Time Number of Nodes
Total Number of Contacts
1 days Day 1: 0 - 342,915 98 4,412,929
Table 4. RollerNet Dataset details.
Duration Start-End Time Number of Nodes
Total Number of Contacts
0.12 day Day 1: 0 - 3096 (51.6 min) 63 80,824
Table 5. RWP_63 Dataset details.
Duration Start-End Time Number of Nodes
Total Number of Contacts
0.12 day Day 1: 0 - 3096 63 576
3.5. Time Window Calculation
The time window is calculated as discussed in Section 3. Evaluated results for INFOCOM’06, RWP_98, RollerNet and RWP_63 is presented in Tables 6-9.
Time window calculation presented in Tables 6-9 re-veals phenomenon of number of time window vs. contact duration per time window. Here, there is concern of total duration and number of nodes participated in simulation.
Temporal algorithm uses the value of Tmin, Tmax, nodes, number of connections and timestamps, time window size as input to compute temporal distance as shown in Tables 10-13.
Table 6. INFOCOM’06 Timewindow calculation.
Tmin (seconds)
Tmax (seconds)
Total Nodes
Number of Connections
Number of Timestamps
Timewindow (w) (seconds)
Day 1 61,260 86,400 96 178,695 8 3240
Day 2 86,400 172,800 98 585,414 26 3240
Day 3 172,800 259,200 93 378,624 27 3240
Day 4 259,200 345,600 83 9227 27 3240
Table 7. RWP_98 Timewindow calculation.
Tmin (seconds)
Tmax (seconds)
Total Nodes
Number of Connections
Number of Timestamps
Timewindow (w) (seconds)
Day 1 0 342,936 98 4,412,929 2598 132
Table 8. RollerNet Timewindow calculation.
Tmin (seconds)
Tmax (seconds)
Total Nodes
Number of Connections
Number of Timestamps
Timewindow (w) (seconds)
Day 1 0 3105 62 2,711,107 207 15
Table 9. RWP_63 Timewindow calculation.
Tmin (seconds)
Tmax (seconds)
Total Nodes
Number of Connections
Number of Timestamps
Timewindow (w) (seconds)
Day 1 0 3124 25 576 44 71
Table 10. INFOCOM’06 static and temporal distance.
Static Distance Average Temporal Distance
Day 1 1.56 0.25
Day 2 1.23 0.51
Day 3 1.3 0.23
Day 4 1.3 1.47
Table 11. RWP_98 static and temporal distance.
Static Distance Average Temporal Distance
Day 1 1.81 0.31
H. SHAH ET AL.
Copyright © 2013 SciRes. CN
268
Table 12. RollerNet static and temporal distance.
Static Distance Average Temporal Distance
Day 1 1.12 0.65
Table 13. RWP_63 static and temporal distance.
Static Distance Average Temporal Distance
Day 1 1.12 0.65
Since our temporal metric presented gives us a better
understanding of the network with respect to the tempor-al dimension since they can provide us an accurate mea- sure of the delay of the information diffusion process that is not possible with traditional static metrics. The values cannot be directly compared but are good indicators for information diffusion.
4. Adaptive Routing In this section we present density aware adaptive routing algorithm that integrates forwarding and replication tech- niques. Here, the threshold is computed based on the average temporal distance gain. It determines sparseness and denseness for given time interval. If the network is dense routing engine uses an encounter based forwarding technique [9], else, it invokes multi period based Spray and Wait [10]. Thus, AR toggles between forwarding and replication technique based on network condition per time window.
4.1. Algorithm Input: minT , maxT , Average temporal distance
Output: Delivery ratio, Overhead ratio Definition: Overhead Ratio = (Relayed Message −
Delivered Message)/Delivered Message 1) Read the values of Tmin, Tmax and average temporal
distance for INFOCOM’06, RollerNet and RWP. 2) Compute the threshold for the Average temporal
distance. At the end of each time stamp calculate the gain of average temporal distance compared to previous time-stamp. Threshold = average of all the gain calculated at the end of each timestamp.
3) If average temporal distance < threshold, then do call encounter based forwarding otherwise do call multi period based Spray and Wait.
4) Generate message state report for message traffic vs. Delivery ratio, message traffic vs. Overhead ratio and message traffic vs. Message dropped.
5) Generate comparison chart based on evaluating re-sults.
4.1.1. Encounter base Forwarding Suppose, node “n” encounters the node “m”, if “n” has a
message for “m” then it sends the message to “m”. After that if “n” has a message for the destination id which is stored in summary vector F [] of “m” then “n” forward the message to “m”. Otherwise, if the encounter rate EV of “m” is greater than the encounter rate of “n” then it for-wards the message to “m”.
F [ ] = vector of frequently encountered nodes V [ ] = message vector which carries a destination id of
the message Wi = current window update interval Upon reception of a Hello message h from node m do ifnewNeighbour(m) == true ifmsgQueue.hasMsgsForDest(m) == true deliverMsgs(m) updateEV() for all destinations d ∈n.V[]do if d∈m.f[] forward message to node m else m.EV>n.EV forward message to node m updateEV() if time ≥ nextUpdate then EV ← α ・CWC + (1 − α) ・EV CWC ← 0 nextUpdate ← time +Wi end if In this, every node maintains encounter value (EV) and
summary vector F[],which stores the frequently encoun-tered node ids. To track a node’s rate of encounter, every node maintains two pieces of local information: EV and current window counter (CWC). EV represents the node’s past rate of encounters as an exponentially weighted moving average, while CWC is used to obtain information about the number of encounters in the current time inter-val. EV is periodically updated to account for the most recent CWC in which rate of encounter information was obtained. Updates to EV are computed as follows:
EV ← α·CWC + (1 − α)·EV This exponentially weighted moving average places an
emphasis proportional to α on the most recent complete CWC. Updating CWC is straightforward: for every en-counter, the CWC is incremented. When the current window update interval has expired, the encounter value is updated and the CWC is reset to zero.
4.1.2. Two Period Spray and Wait It is based on multi period Spray and Wait [11].The algo-rithm starts with spraying fewer message copies than the copies decided as above, and then waits for a certain pe-riod of time to see if the message is delivered. When the delivery does not happen, the algorithm sprays additional copies of a message next period, and again waits for the delivery.
Sprays L1 copies to the network at the beginning of
H. SHAH ET AL.
Copyright © 2013 SciRes. CN
269
execution and additional L2-L1 copies at next time inter-val.
TwoPeriodsw(L, α, td ) optcost = L; for each 0 < L1 < L do L2floor = max[ L+1, L1+(α L1 td (L- L1))]; for L2 = L2floor , L2floor +1 do if c2(L1, L2) < opt_cost then opt_cost = c2(L1, L2); opt_cts = [L1, L2] end if end for end for return opt_cts
where, td = Time deadline L = Number of message copies α = 0.0015*number of copies L1 = Number of message copies to spray in the first time
period L2 = Number of message copies to spray in the second
time period c2 = Cost as number of copies used per message
4.2. Simulation setup and Performance Analysis In our experiments, ONE (Opportunistic Network Envi-ronment) [12] simulator is used to evaluate the protocols. This scenario consists of a 4500 × 3400 m area of Hel-sinki city. These nodes are divided into 6 groups. Group 1 and Group 3 are pedestrian group. Group 2 is automo-bile group and Group 4, Group 5 and Group 6 are trolley- bus group. Pedestrians move with speeds of 0.5 - 1.5 m/s and with waiting time of 0 - 120 seconds. Automobiles move with speeds of 10 - 15 km/h and with waiting time of 0 - 120 seconds. Trolleybuses move with speeds of 7 - 10 km/h and with waiting time of 10 - 30 seconds at every station. Two kinds of devices are used in the expe-riments. One is Bluetooth device which transmission range is 10 m and transmission speed is 2 Mbps. The other is 802.11b WLAN device which transmission range is 30 m and transmission speed is 4.5 Mbps. The buffer of these mobile devices is up to 1 MB. When buffer is full, new messages will not be accepted by the node until some old messages are deleted from the buffer. The dura-tion of the simulation is 43,000 simulation seconds.
Assumptions It is assumed that the network is congestion free and
node have infinite energy. Default buffer management policies are used. No explicit settings made to schedule
and drop policies. Datasets Real and Synthetic datasets are used for simulation.
For, the real trace file INFOCOM’06 and RollerNet trace files are used. For synthetic data RWP model is used. Information about the name, location, duration, partici-pants and address IDs are summarized in Tables 2-5. First temporal distance of real traces and synthetic data-sets are evaluated and presented in Tables 10-13. Real trace are first converted to ONE compatible format and then imported for simulation. RWP dataset is generated using ONE simulator.
Performance Metrics Delivery Probability: The delivery probability is a mea-
sure of the fraction of the created packets that are deli-vered to the destination. This is the ratio of the total number of packets that are delivered to their destination to the total number of packets that are created.
Overhead Ratio: The overhead ratio is calculated using the following equation.
Here, the term relayed messages refers to the messages that have been forwarded from the source to an interme-diate node to be forwarded towards the destination. This number is a measure for the number of packets or copies of packets that have been inducted into the network. The number of delivering messages refers to the total number of creating packets that are successfully delivered to the destination.
INFOCOM’06 Delivery ratio, overhead ratio and number of dropped
messages comparison for AR, Epidemic and PRoPHeT routing with INFOCOM’06.
As shown in Figure 2 the delivery ratio of AR ob-served better compared to Epidemic and ProPHeT. But, overall value is still poor. This is due to INFOCOM 2006 had a smaller number of timestamps and larger time window size compared to RWP. Hence, AR with INFO-COM’06 executed forwarding and replication all most same number of times.
Figure 3 indicates that AR has a low overhead ratio due almost the same number of time forwarding and rep-lication executed. Thus, in forwarding numbers of trans-missions are less and resulting in a limited message compared to the Epidemic and ProPHeT. Where in Epi-demic is purely replication and ProPHeT is probability based multi copy scheme.
Number of message dropped in AR is low compared to ProPHeT and Epidemic. Since, in forwarding phase node’s encounter rate checked and in replication two
Overhead Ratio
Number of relayed messages Number of delivered messagesNumber of delivered messages
−=
H. SHAH ET AL.
Copyright © 2013 SciRes. CN
270
period Spray and Wait used. This resulted in dropping the less number of messages as shown in Figure 4.
RollerNet As shown in Figure 5 RollerNet traces have the num-
ber of connections. Thus created more contact opportuni-ties. AR calls more number of times forwarding which result lesser message drop and better delivery ratio.
In a RollerNet numbers of connections are high com-pared to RWP. Thus, the network is dense and AR fre-quently executed forwarding technique. This resulted in lower overhead as presented in Figure 6.
AR in forwarding phase uses encounter based tech-nique. Thus due to higher contacts opportunities forwar- ding is frequently called. Hence, the numbers of dropped
Figure 2. Delivery Ratio Comparison of AR, Epidemic and ProPHeT.
Figure 3. Overhead Ratio Comparison of AR, Epidemic and ProPHeT.
Figure 4. Number of Dropped Messages Comparison of AR, Epidemic and ProPHeT.
messages are zero and results in minimum overhead as shown in Figure 7.
5. Conclusion and Future Work It reveals that the node mobility plays a vital role in the efficient diffusion of information in challenging envi- ronment. And while doing so one cannot ignore to under-stand the movement patterns and related properties such as time order, frequency, contact duration, inter-contact time, etc. These properties are applied in designing an efficient routing in IC-MANET. And understanding the dynamics of the network and thereby taking forwarding
Figure 5. Delivery Ratio Comparison of AR, Epidemic and ProPHeT.
Figure 6. Overhead Ratio Comparison of AR, Epidemic and ProPHeT.
Figure 7. Number of Dropped Messages Comparison of AR, Epidemic and ProPHeT.
H. SHAH ET AL.
Copyright © 2013 SciRes. CN
271
or replication decisions. It is established that routing effi-ciency terms of delivery ratio and overhead are signifi-cantly improved with temporal metrics. In future, it can be extended for contact sequence based probabilistic routing and temporal closeness centrality based evalua-tions.
6. Acknowledgements We express our sincere gratitude to the management of Ganpat University—Mehsana and Marwadi Education Foundation—Rajkot; for providing us research opportun-ities and their wholehearted support for such activities.
REFERENCES [1] T. Spyropoulos, K. Psounis and C. S. Raghavendra, “Sin-
gle-Copy Routing in Intermittently Connected Mobile Networks,” The First Annual IEEE Communications So- ciety Conference on Sensor and Ad Hoc Communications and Networks, 4-7 October 2004, pp. 235-244.
[2] K. Fall, “A Delay-Tolerant Network Architecture for Challenged Internets,” Proceedings of the 2003 Confer- ence on Applications, Technologies, Architectures, and Protocols for Computer Communications, 2003, pp. 27- 34.
[3] V. Kostakos, “Temporal Graphs,” Physica A: Statistical Mechanics and Its Applications, Vol. 388, No. 6, 2009, pp. 1007-1023.
[4] J. Tang, M. Musolesi, C. Mascolo and V. Latora, “Char- acterising Temporal Distance and Reachability in Mobile and Online Social Networks,” ACM SIGCOMM Com- puter Communication Review, Vol. 40, No. 1, 2010, pp. 118-124. doi:10.1145/1672308.1672329
[5] E. Nordström, P. Gunningberg and C. Rohner, “Haggle: A Data-Centric Network Architecture for Mobile De-
vices,” Proceedings of the 2009 MobiHoc S3 Workshop on MobiHoc S3, 2009, pp. 37-40.
[6] P.-U. Tournoux, J. Leguay, F. Benbadis, J. Whitbeck, V. Conan and M. Dias de Amorim, “Density-Aware Routing in Highly Dynamic DTNs: The RollerNet Case,” IEEE Transactions on Mobile Computing, Vol. 10, No. 12, 2011, pp. 1755-1768. doi:10.1109/TMC.2010.247
[7] F. Bai and A. Helmy, “IMPORTANT: A Framework to Systematically Analyze the Impact of Mobility on Per- formance of Routing Protocols for Adhoc Networks,” 22nd Annual Joint Conference of the IEEE Computer and Communications Societies, Vol. 2, 30 March 2003-3 April 2003, pp. 825-835.
[8] J. Tang, M. Musolesi, C. Mascolo and V. Latora, “Tem- poral Distance Metrics for Social Network Analysis,” Proceedings of the 2nd ACM workshop on Online Social Networks, 2009, pp. 31-36.
[9] J. Zhang, G. Luo, K. Qin and H. Sun, “Encounter-Based Routing in Delay Tolerant Networks,” International Con- ference on Computational Problem-Solving (ICCP), Cheng- du, 21-23 October 2011, pp. 338-341. doi:10.1109/ICCPS.2011.6089802
[10] E. Bulut, Z. Wang and B. K. Szymanski, “Cost Efficient Erasure Coding Based Routing in Delay Tolerant Net- works,” IEEE International Conference on Communica- tions, 2010, pp. 1-5. doi:10.1109/ICC.2010.5502382
[11] E. Bulut, Z. Wang, and B. K. Szymanski, “Cost-Effective Multiperiod Spraying for Routing in Delay-Tolerant Net- works,” IEEE/ACM Transactions on Networking, Vol. 18, No. 5, 2010, pp. 1530-1543. doi:10.1109/TNET.2010.2043744
[12] A. Keränen, T. Kärkkäinen and J. Ott, “Simulating Mo- bility and DTNs with the ONE (Invited Paper),” Journal of Communications, Vol. 5, No. 2, 2010, pp. 92-105.