Content uploaded by Sujata Pal
Author content
All content in this area was uploaded by Sujata Pal on Sep 08, 2015
Content may be subject to copyright.
1
Distributed Information-Based Cooperative
Strategy Adaptation in Opportunistic Mobile
Networks
Sudip Misra, Senior Member, IEEE, Sujata Pal, Member, IEEE, Barun Kumar Saha
Email: {smisra, sujata.pal, barun.kumar.saha}@sit.iitkgp.ernet.in
Abstract—Cooperation among nodes is a fundamental necessity in Opportunistic Mobile Networks (OMNs), where the messages are
transferred using the store-carry-and-forward mechanism, due to sporadic inter-node wireless connectivity. While multiple works have
addressed this issue, they are often constrained in their assumptions on solutions (e.g., requirement of central authority, and tracing
the recipient nodes for providing reward or punishment). In this work, we address this research lacuna by taking an evolutionary theory-
based approach. In evolutionary theory, the players analyze alternative strategies and select the best one to survive in a population.
Inspired by this, in this work, we propose a Distributed Information-Based Cooperation Ushering Scheme (DISCUSS) to promote
cooperation in message forwarding between nodes. In this scheme, the nodes maintain and exchange information with one another
during contacts about the messages created and delivered in the network. Based on this, the nodes evaluate their own performance and
compare that with the approximated network performance to adapt the most successful forwarding strategy. Simulation results show
that the message delivery ratio in the network improves upto 31%, when the nodes dynamically switch their strategies, as compared to
the case when they do not. Furthermore, the DISCUSS scheme fared closely to its variant with the nodes having complete knowledge
about the network-wide performance.
Index Terms—Cooperation, Opportunistic Mobile Networks, Strategy adaptation.
F
1 INTRODUCTION
Opportunistic Mobile Networks (OMNs) are a sub-
class of delay-tolerant networks (DTNs) [1]. In tra-
ditional wireless networks, end-to-end communication
paths usually exist between the source and destination
node pairs, which, however, is not true for the OMNs.
The messages in OMNs are “opportunistically” routed
to their destinations using the store-carry-and-forward
transfer mechanism. Mobility among users creates new
opportunities to connect and communicate with one
another using wireless devices [2]. Since the source-
destination pairs might not be connected simultaneously,
route maintenance exhibits a huge challenge. So, the
routes are typically computed at each node based on
the local knowledge of the node. A source node delivers
its message to the destination node, if both the nodes
come in contact of each other. Otherwise, the message is
forwarded to the intermediate nodes, which then store
and carry the messages, until one of them meets with the
destination and delivers, or they can find other nodes
which can do the same. Thus, message forwarding in
OMNs significantly depends on the cooperation of the
intermediate nodes.
Existing routing protocols such as [3]–[7] make the as-
sumption that the nodes are fully cooperative. Since the
•Sudip Misra, Sujata Pal, and Barun Kumar Saha are with the School of
Information Technology, Indian Institute of Technology Kharagpur, India.
E-mail: {smisra, sujata.pal, barun.kumar.saha}@sit.iitkgp.ernet.in
nodes are typically unmanaged, it is difficult to coordi-
nate different nodes having different goals. So, we do not
always consider fully cooperative behavior among all
nodes. Mobility in OMNs causes nodes to meet oppor-
tunistically. These opportunistic connections help certain
nodes to transfer their messages through intermediate
nodes to the corresponding destination nodes. Some
nodes may take help from others for forwarding their
own messages, but may not always help in forwarding
others’ messages. This selfish behavior heavily affects
the network performance. The mobility and the greedy
nature of the nodes require a distributed and efficient
cooperation scheme that can stimulate cooperation and
discourage non-cooperative behavior.
Enforcing cooperation among nodes has been exten-
sively studied in the literature. One way of promoting
cooperation among nodes is to give reward or pun-
ishment. However, for giving reward or punishment,
the recipient node needs to be traced. Zhu et al. [8]
proposed ‘iTRUST’, in which a periodically available
trusted authority judges each node and detects the
misbehaving nodes. Credit-based incentive schemes [9]–
[11] were proposed, in which credit is given to the
forwarding nodes in terms of virtual pricing unit. Wei et
al. [12], and Zhang et al. [13] proposed reputation-based
incentive schemes for encouraging cooperation among
nodes. Wu et al. [14] proposed a game theoretic approach
that encourages cooperation among nodes. However, the
scheme proposed by these authors is based on the use
of a server (Credit Clearance Centre) for updating each
Accepted Version (For Personal Use Only)
2
transaction. In another scheme, ‘Pi’ [15], the source node
attaches some incentives on the forwarding message.
In this work, we consider three groups of nodes based
on three strategies of message forwarding – cooperate,
exploit, and isolate – similar to [16]. The cooperators help
the other nodes in forwarding their messages. The ex-
ploiters use the other nodes as “free-riders”, without
helping them in forwarding their messages. The isolators
neither take help, nor provide so, to the other nodes in
forwarding their messages. It is pertinent to clarify at
this juncture that the existing work [16] considers the
above mentioned strategies. It, however, merely shows
the existence of a cyclic equilibrium among the three
strategies, cooperate, exploit, and isolate. Additionally, it
studies the aspect of cooperation in opportunistic mobile
networks, combining the concept of Rock-Scissors-Paper
game, and Simpson’s paradox. The current work, on
the contrary, presents a distributed mechanism, using
which each node dynamically selects the most successful
strategy. Our mechanism prevents the nodes from se-
lecting the exploit and isolate strategies, which, in turn,
stimulates them to cooperate. The design of a mecha-
nism for the enforcement of cooperation is new to this
work. Further, in this work, we present the theoretical
characterization of the proposed scheme together with
extensive performance evaluations.
Evolutionary theory [17] postulates that each individ-
ual in a population periodically checks the alternative
strategies and selects the best one for survival. Inspired
by this, we consider an OMN, in which the nodes peri-
odically compare their individual performance with the
locally available network-wide performance (in terms of
delivery ratio of messages) and switch to the strategy
of the most successful group, if it is not the node’s
strategy already. We develop a Distributed Information-
baSed Cooperation UShering Scheme (DISCUSS), using
which the nodes locally exchange information among
themselves to acquire knowledge about the performance
of one another, without requiring a central authority
to do it. We show that the proposed scheme provides
high delivery probability and low latency on message
delivery. For evaluating the effectiveness of the proposed
scheme, we proposed a variant of DISCUSS, in which the
nodes are assumed to possess global knowledge – every
node is assumed to possess precise information about the
messages created and delivered by any node. In the rest
of the manuscript, we refer to this variant as “DISCUSS
with global knowledge”.
The contributions of this paper are summarized in the
following points.
•Proposing a scheme, inspired by evolutionary the-
ory, following which the nodes dynamically adapt
their message forwarding strategies based on dis-
tributed information over time.
•Proposing the distributed information based co-
operation ushering scheme, DISCUSS, to promote
cooperation amongst nodes for message forwarding
through self-evaluation using their own and the
network’s performance information.
•Establishing the effectiveness of DISCUSS by evalu-
ating the Jaccard similarity index [18] between DIS-
CUSS and its variant, in which global knowledge is
available.
•Analyzing the theoretical characteristics of the
scheme in terms of complexity and convergence.
•Demonstrating the efficiency of the proposed
scheme through extensive simulations using both
real-life traces and synthetic mobility models.
The remaining part of this work is organized as fol-
lows. Section 2 presents a review of the related work.
Section 3 provides an overview of the system model.
This is followed by Section 4, in which we describe
the proposed scheme, DISCUSS. Section 5 contains the
theoretical description of the characteristics of DISCUSS.
Sections 6 and 7 present the simulation setup and analy-
sis of results. Finally, in Section 8, we conclude this work.
2 RE LATED WORK
The operation of message delivery in OMNs is gen-
erally reliant upon cooperation by intermediate nodes.
Due to resource limitations, the intermediate nodes do
not always want to help the other nodes. Each node
in the network always tries to fulfill its own goal by
utilizing the other node’s resources. So, to promote
cooperation between nodes in such type of networks,
different cooperation-enforcing schemes [19] were pro-
posed. Credit- and reputation-based approaches [9]–[13]
are commonly known to promote cooperation among
nodes. In the credit-based incentive schemes, virtual
currency or pricing acts as the credit. On the other hand,
in the reputation-based schemes, reputation of the nodes
are calculated by their neighbors based on the message
forwarding actions.
2.1 Credit-based Schemes
Shevade et al. [10] first proposed a credit-based incentive
mechanism for DTNs, which motivate the selfish nodes
to cooperation. Their results show that the presence
of selfish nodes degrade the overall network perfor-
mance. To alleviate this behavior, they proposed the
pair-wise Tit-for-Tat (TFT) incentive mechanism. They
considered two constraints – generosity and contrition
– to maximize cooperation among nodes. Zhu et al.
[9] proposed SMART, a Secure Multilayer Credit based
Incentive scheme for DTNs. SMART stimulates cooper-
ation among nodes by preventing the malicious users
from cheating credits based on layered coins. Layered
coin provides virtual credits for charging and rewarding
of data forwarding in DTNs. On the other hand, another
scheme, MobiCent [11], provides incentive to the selfish
nodes for forwarding other node’s messages. Lu et al.
[15] proposed the “Pi” protocol, where the selfish nodes
are stimulated to cooperate by forwarding the messages
of other nodes. Pi attaches incentives to the messages
Accepted Version (For Personal Use Only)
Accepted Version (For Personal Use Only)
3
and sends it to the intermediate nodes, which attracts
the participation of other nodes in forwarding. However,
this method need a Trusted Authority (TA) for storing
the credit and reputation of each nodes. “iTrust” [8]
encourages cooperation among the forwarding interme-
diate nodes, by providing incentives, and detecting and
punishing the misbehaving nodes. “iTrust” introduces
the concept of a periodically accessible trusted author-
ity, which detects the misbehaving nodes based on the
forwarding history evidence.
2.2 Reputation-based Schemes
MobiID [12] is a reputation-based incentive scheme. It
assumes an offline system manager, where each node
registers to join the network. This scheme uses self-check
and community-check of either the previous-/next-hop
nodes or their community before forwarding a message.
A reputation value is assigned to the forwarding node,
when it forwards a message to another node. Nodes
are ostracized when their corresponding reputations are
less than a given threshold. So, each node exerts itself
to cooperate in forwarding other node’s messages, in
order to earn increased levels of reputation. Zhang et
al. [13] proposed a practical reputation-based incentive
“Pri” scheme. Pri also assumes the use of Offline Security
Manager (OSM), which is in charge of key distribution.
Nodes register in OSM before joining the network.
Resta and Santi [20] analyzed the performance of
epidemic routing for different degrees of node coopera-
tion. They derived the probability distribution of packet
delivery delay and communication cost. Communication
cost is computed by determining the number of nodes
having the message in the network by the timing a
copy of the message is first delivered to the destination.
Apart of these works, Buttyan et al. [21] and Yin et al.
[22] proposed a non-cooperative game theoretic model
that avoids selfish behavior among nodes and stimulates
cooperation among individuals in a network. Elwhishi et
al. [23] proposed a message scheduling framework for
epidemic routing in DTNs, which improves the message
delivery ratio.
From the above works, it can be synthesized that co-
operation enforcement poses a great challenge in OMNs
and requires the use of tamper-proof hardware for up-
dating the reputation values of every node. Otherwise,
for giving reward or punishment, the recipient nodes
need to be traced, which is costlier in OMNs. On the
other hand, the proposed approach does not need a
central server or any incentive mechanism. On every
contact, a node maintains and exchanges information
with the other neighboring nodes, about the messages
created and delivered in the network. Based on this
local information, a node evaluates its own performance
and compares it with the approximated network perfor-
mance of others. A node adapts, other strategy, if its own
performance is inferior than the others performance.
3 SYSTEM MODEL
3.1 Assumptions
The proposed schemes “DISCUSS” and its variant “DIS-
CUSS with global knowledge”, makes the following
assumptions.
1) Each node follows any one of these three message
forwarding strategies at a time – cooperate,exploit,
and isolate.
2) In the DISCUSS scheme, the nodes reliably ex-
change information with their neighboring nodes.
3) In case of DISCUSS with global knowledge, the
nodes can obtain message delivery information
from a Central Authority (CA).
4) In DISCUSS with global knowledge, every node
shares its relevant information with the CA. So,
each node has “complete” information (number of
generated and delivered messages) about the other
nodes.
3.2 Strategy Defined-Opportunistic Mobile Network
We represent an OMN as (N, M, S ), where Ndenotes
the set of nodes in the network, Mdenotes the set of
messages generated by the nodes, and Sdenotes the
set of strategies selected by any node in forwarding
the messages; S={Cooperate, Exploit, I solate}. Each
node may act as a source, a destination or an interme-
diate relay node. We consider three groups of nodes
–cooperators,exploiters, and isolators – based on these
strategies. The following section elaborates the behavior
of the individual nodes.
3.2.1 Cooperators
These nodes with the strategy “cooperate” not only
forward their own messages, but help the other nodes as
well in doing so. In other words, the cooperators act as
relays by receiving, storing and forwarding the messages
generated by the other nodes.
3.2.2 Exploiters
On the other hand, the exploiters forward their messages
to the other nodes (cooperators) for delivery. They re-
ceive other node’s messages, but instead of storing them,
they silently drop those messages. The exploiters take
help from others for forwarding their own messages as
free riders, without helping them.
3.2.3 Isolators
The isolators only receive the message for which they are
the destinations. They do not take help from the other
nodes for forwarding their messages, neither do so for
others. The isolators directly deliver their messages, when
they meet with the corresponding destination node.
Definition 1: An OMN is said to be Strategy Defined-
Opportunistic Mobile Network (SD-OMN), if all the nodes
in the OMN follow any message forwarding strategy
described by the set S={Cooperate, Exploit, I solate}.
Accepted Version (For Personal Use Only)
4
TABLE 1: List of Notations
CM i
tCreated messages list of the node iat time t
DM i
tDelivered messages list of the node iat time t
DP i
tDelivery probability list of the node iat time t
Ψg
Accepted Version (For Personal Use Only)
Group weight of the group g
Ψi(t)Group weight of the node iat time t
pi
kDelivery probability of kth node of node i
tkTime stamp of kth node
αi
gDelivery probability of the group g,
g∈ {C, E, I }, as known by the node i
ωi
gWeighted factor of the group g,g∈ {C, E, I },
as computed by the node i
γi
gWeighted average delivery ratio of group g,
g∈ {C, E, I }, as computed by the node i
At any point of time, a given node chooses only one
strategy from S.
Let nC,nE, and nI, respectively denote the number of
cooperators, exploiters, and isolators in the SD-OMN at
time instant t. Therefore, we have |N|=nC+nE+nI. The
objective of SD-OMN is to motivate the non-cooperators
in the network into cooperation in order to optimize the
performance of the network. Therefore, the goal function
can be write as,
max
t∈RnC(t),s.t. 0< t ≤Γ,0< nC(t)≤ |N|(1)
where, Γis the lifetime of the system. When the con-
cerned performance objective is the delivery ratio (i.e.
the fraction of the created messages that have been deliv-
ered to their respective destinations) of the messages, the
performance function should be maximized. Therefore,
in this context, the goal function of this work, can be
alternatively represented as,
max
t∈RX
s∈S
αs(t),s.t. 0< t ≤Γ,0< αs(t)≤1(2)
where, αs(t)denotes the delivery ratio of the group of
nodes with strategy s∈Sat any time instant t. We
validate this in Section V.
3.3 Network Performance Information
Initially, each node has a message forwarding strategy.
Each node distributively shares information with the
other nodes, without requiring a CA. Each node has
limited information about the effect of its actions. When
two nodes mutually encounter, they interchange their
past history (own delivered message list and nodes’
delivery probability list) with the neighboring nodes.
Based on the local knowledge of past history, each
node updates its behavior. The nodes select their own
communication mode – whether they cooperate,defect or
isolate in message forwarding, based on their past history.
In case of DISCUSS with global knowledge, we assume
that all nodes are managed by a CA. The nodes share
their relevant information such as received message ids,
message sender node ids, and own delivery probability
with the CA. Every node collects information such as
delivery probabilities of the other nodes’ from the CA.
They select their own communication modes based on
these information.
4 DISTRIBUTED INFORMATION-BASED COOP-
ERATION USHERING SCHEME
We now discuss a scenario where a node dynamically
switches its forwarding strategy, if required, to the most
successful strategy in the concerned SD-OMN. This is
inspired by the existing research in evolutionary theory
(ET) [17], which shows that the size of groups in a pop-
ulation1changes with the various competing strategies.
In particular, the following presents a mapping of the
concepts from Darwin’s Natural Selection Theory [24]
into our context involving SD-OMN.
•Nodes in SD-OMN exhibit variation in their mes-
sage forwarding actions.
•Such variation occurs as the nodes follow different
kinds of strategies, as defined by the set S.
•The most successful strategy among the set Ssur-
vives in the population and is propagated to the
other nodes over time.
We present the proposed scheme, DISCUSS, in the back-
drop of ET. Table 1 list the notations used to represent
the system parameters with a short description.
DISCUSS comprises of two phases that are repeated in
every generation interval (τ). (1) Acquiring information on
the performance of the SD-OMN, and (2) Strategy adap-
tation. The first phase requires information on – (a) The
messages created (CM ) by the nodes, (b) The delivered
messages (DM ), and (c) The delivery probabilities (DP )
of the nodes.
4.1 Acquiring Information
Let iand jbe two nodes that come in contact with
each other at time t. Let CMi(t)be the list of generated
messages by node iat time t. Then,
CM i(t) = {(mi
r, tr)},
where mi
ris the rth message created by the node i, at
time tr,r≥0. In other words, CM i(t)consists of the list
of messages created by node itill time t. Similarly, we
define DM i(t)as:
DM i(t) = {(mi
q, tq)},
where mi
qis the qth delivered message of the node iat
time tq. Therefore, DMi(t)⊆CM i(t).
Each node imaintains its own delivery probability, as
well as its view of the other nodes’ delivery probabilities
known till time tin DPi
tas:
DP i
t={(k, pi
k, ti
k)}, k = 1 to N, t ≥tk,∀k
1. Population in ET and set of nodes in our case
5
Here, pi
kis the last updated information stored at node
ion the delivery probability of the node kknown at time
tk. The delivery probability (pk) of any node k,∀k∈N,
is determined as:
pk=|DM k(t)|
|CM k(t)|(3)
Information are updated between pairs of nodes,
when they come within their respective communication
ranges. More specifically, node i,i∈N, maintains a list
of delivery probabilities for every node it has met before.
Delivery probability (pi
k) of the list DP i
tis updated in
every connection, according to the following rule:
pi
k=(pj
k,if tj
k> ti
k,
pi
k,otherwise. (4)
Algorithm 1 describes the steps required for updating
node i’s lists on encountering node j,∀i, j.
Algorithm 1: Delivery probability update by node i
on contact with node j
Inputs :CM i
t, CM j
t, DM i
t, DM j
t, DP i
t, DP j
t
Output: Updated values of DM i
tand DP i
t
1C← |CM i
t|;// Number of messages in CM i
2D← |DM j
t|;// Number of messages in DM j
3/*Check, whether DM jlist contain any
created messages of node i. If so, add
it to DM i.*/
4for u←1, C do
5for v←1, D do
6if (mu=mv)then
7Add mvto DM i
t;
8end
9end
10 end
11 // Find delivery probability of iand j
12 DP i
t← |DM i
t|/|CM i
t|;
13 DP j
t← |DM j
t|/|CM j
t|;
14 Pi← |DP i
t|;Pj← |DP j
t|;
15 /*Node icompare its own DP list with
node j’s DP list */
16 for u←1, Pido
17 for v←1, Pjdo
18 if tv
k> tu
kthen /*Compare time stamp
of kth node at node iand the same
kth node at node j.*/
Accepted Version (For Personal Use Only)
19 pu
k←pv
k, tu
k←tv
k;/*Update delivery
probability and time stamp of
kth node at node i*/
20 end
21 end
22 end
Figure 1 illustrates an example of comparison and
update process of two nodes iand j. On every contact,
node iupdates its information with node j, as given
Fig. 1: Update information on each contact
below.
•Node ihas three lists CM i,DM i, and D P i, at time
t. Similarly, node jhas CM j,DM j, and D P j.
•Node igenerates two messages, M8and M12, at
t= 150sand 200s, respectively. Similarly, node j
has messages M24 and M5created at t= 321sand
101s, respectively.
•Nodes iand jlist the past delivered messages
of nodes. Node ichecks node j’s list of deliv-
ered message, and updates its own list of de-
livered messages. Initially, nodes iand jhave
DM i={hM8,400i,hM2,700i} and DM j=
{hM24,800i}. After updating, node ihas DM i=
{hM8,400i,hM2,700i,hM24,800i}.
•Node ifinds its delivery probability (pi) at current
time tas pi=|CM i|/|DMi|, and updates its deliv-
ery probability list.
•Node icompares its own DP ilist with the jth
node’s DP jlist, and then updates its list.
•Let node ichoose the kth node time stamp (ti
k) from
the DP ilist, and compare it with the kth node’s
time stamp (tj
k) of node j’s DP jlist. If ti
k< tj
k,
node iupdates its kth node’s delivery probability
by dpj
k. Node ichecks the 1st node’s time instant,
i.e. 300 >0, and consequently makes no change.
Again, it checks the 2nd node’s time instant, i.e.,
400 <800, and then updates the message delivery
probability of this node by 0.5. Similarly, it repeats
this for the rest of the nodes of node i. Hence,
DP i
t={h0.2,300i,h0.5,800i,h0.3,600i}.
Algorithm 1 repeats exactly in the same way as described
for updating the list of other nodes.
Further, each node i,i∈N, maintains a record of
the group weights,Ψi
g(t),g∈ {C, E, I }. Initially, at time
t= 0,Ψi
g(t)is empty. At any time t, when a node i
comes in contact with another node jbelonging to a
group g, where g∈ {C, E, I }, node iupdates its value
of the corresponding group weight as:
Ψi
g(t)=Ψi
g(t−1) + Ii
g(t),(5)
where, Ii
g(t)is the indicator decision variable, which, in
turn, is defined as follows:
Ii
g(t) = (1,if node ireceives a message from group g,
0,otherwise.
6
For example, Ψi(t) = {Ψi
C,Ψi
E,Ψi
I}indicates that until
the time instant t, node ireceives Ψi
C,Ψi
Eand Ψi
I
number of messages, respectively, from the cooperators,
exploiters, and isolators, respectively. Thus, Ψgdenotes
the relative importance of each group over the others.
4.2 Strategy Adaptation
Cooperation is necessary and important in OMNs for
increasing the performance of the networks. So, for
enhancing cooperation among nodes, we propose a dy-
namic strategy adaptation mechanism. This mechanism
bridges the nodes’ self-recommendation approach based
on the local information gathered by a node during
contacts with other nodes. The objective of the proposed
approach is to maximize the expected delivery proba-
bility of messages in the Kth time-slots by motivating
the exploiters and isolators to shift their strategies to
cooperation.
We evaluate the proposed approach using two scenar-
ios: static and dynamic. In the static scenario, the number
of cooperators (nC), exploiters (nE), and isolators (nI)
remain invariant over time t, i.e.,
nC(t) = nC(0),
nE(t) = nE(0),
nI(t) = nI(0),
(6)
and nC+nE+nI=|N|,∀t.
In the dynamic scenario, the number of cooperators
(nC), exploiters (nE), and isolators (nI) vary with time.
Ageneration is a stage in the life cycle of a node. The
strategies of individuals (nodes) remain same in a genera-
tion. The generation interval (τ), where τ={1,2,3, ..., K},
is the time required to complete a generation. In Fig-
ure 2, initially, the size of cooperators, exploiters and
isolators are the same. After a generation interval, every
Fig. 2: Strategy adaptation by the nodes in the SD-OMN
after a generation interval (τ).
node compares its individual performance with that of
the other groups, and dynamically changes its strategy,
if required, accordingly. Such a self–evaluation can be
performed by each node individually during the run-
time based on information exchanged with the other
nodes, as discussed below.
Each node ihas a delivery probability list (DP i
t),
where it stores the delivery probabilities of the other
nodes based on the most recent information available,
together with the corresponding time when the infor-
mation was obtained, and then time-stamps it. Any
node i,i∈N, computes the average delivery ratio αg,
g∈ {C, E, I }of cooperators, exploiters, and isolators, as
follows:
αi
C=Pk∈NCpi
k
|NC|
αi
E=Pk∈NEpi
k
|NE|
αi
I=Pk∈NIpi
k
|NI|
(7)
where, αC,αE, and αI, respectively, denote the average
delivery ratio of the cooperators, exploiters, and isolators
in the OMN. Here, NC,NE, and NIdenote the set of
cooperators, exploiters, and isolators that node i,i∈N,
met within time interval τ.
Definition 2: The weighted factor, ω, of a group g,g∈
{C, E, I }is defined as:
ωg=Ψg
Ps∈{C,E,I }Ψs
Node idetermines the weighted factor of each group from
Ψi(t). The weighted factor of cooperators (ωi
C), exploiters
(ωi
E), and isolators (ωi
I) of node iare calculated using
Definition 2, as follows:
ωi
C=Ψi
C
Ψi
C+ Ψi
E+ Ψi
I
ωi
E=Ψi
E
Ψi
C+ Ψi
E+ Ψi
I
ωi
I=Ψi
I
Ψi
C+ Ψi
E+ Ψi
I
(8)
Definition 3: The weighted average message delivery
ratio γgof a group gis defined as:
γg=αg×ωg.(9)
From Definition 3, the weighted average delivery ratio of
cooperators (γC), exploiters (γE), and isolators (γI) of
node iis calculated as follows:
γi
C=αi
C×ωi
C
γi
E=αi
E×ωi
E
γi
I=αi
I×ωi
I
(10)
Finally, node idetermines the most successful group,
gsucc, as follows:
gsucc =
C, γC> γE, γC> γI
E, γE> γC, γE> γI
I, γI> γE, γI> γC
(11)
Now, node ichanges its group to gsucc , if its delivery
probability, computed using Equation 3, is less than the
maximum of the weighted average delivery ratio of the
Accepted Version (For Personal Use Only)
7
three groups, i.e.,
pi< max(γi
C, γi
E, γi
I)(12)
The Strategy Adaptation mechanism is repeated exactly
in the same way for every node after each generation
interval for switching to a successful strategy. In par-
ticular, we note that, if a node fails to perform self-
evaluation due to one reason or another (for example,
when a node does not have information about the other
nodes available yet), it continues with its own previous
strategy. The flowchart presented in Figure 3 shows the
sequence of steps performed by a node executing the
DISCUSS algorithm in an OMN.
Fig. 3: Flowchart depicting the steps in DISCUSS.
5 CHARACTERISTICS OF DISCUSS
This Section presents the complexity analysis and theo-
retical analysis on the characteristic of DISCUSS.
5.1 Theoretical Analysis
Definition 4: A SD-OMN with Nnodes is said to have
reached an equilibrium, when (1−δ)N(where 0≤δ < 1)
nodes have a common strategy, s∈S, and the remaining
δN nodes have different strategy/strategies. The com-
mon strategy of the (1 −δ)Nnodes is said to be the
equilibrium strategy.
Definition 5: A SD-OMN is said to have reached (K, δ)
convergence, if, for Ksuccessive generations, where K≥
1, at least the (1 −δ)Nnodes have the same common
strategy, s∈S, while the remaining nodes could possibly
have different strategy/strategies.
Theorem 1: Necessary criteria for (K, δ)convergence: Let
sibe the equilibrium strategy at the ith generation. If
si=sj,∀i,j∈ {1,2, ..., K}, SD-OMN has attains (K, δ)
convergence.
Proof: The proof follows from Definition 5.
Theorem 2: Sufficient criteria for (K, δ)convergence: Let
sibe the equilibrium strategy for the ith generation. It
is sufficient to say that SD-OMN moves toward (K, δ)
convergence with δ→0, when,
γsimax
si
{γs0
i}, s0
i∈S− {si}
Proof: Let Nsibe the set of nodes with strategy si,
and Ns0
ibe the set with strategies other than si. Using
Equation (9), we have,
γsi=X
n∈Nsi
αn
siωn
si
Assuming, γsiγs0
i, we have
X
n∈Nsi
αn
siωn
siX
n∈Ns0
i
αn
s0
i
ωn
s0
i
⇒X
n∈Nsi
αn
siΨn
siX
n∈Ns0
i
αn
s0
i
Ψn
s0
i
⇒X
n∈Nsi
pn
si
|Nsi|Ψn
siX
n∈Ns0
i
pn
s0
i
|Ns0
i|Ψn
s0
i
The above holds true if any one of the following con-
ditions holds: (1) |Nsi||Ns0
i|, (2) individual αsiΨsi
αsi
0Ψsi
0, (3) both (1) and (2). Let us look at the individual
cases.
Case 1:
|Nsi||Ns0
i|
⇒|Nsi|
|Ns0
i
|1
⇒|Nsi|+|Ns0
i
|
|Ns0
i
|1
⇒1
δ1⇒δ1⇒δ→0.
This implies that SD-OMN moves towards convergence.
Case 2: αsiΨsiαsi
0Ψsi
0implies that most of the nodes
from Ns0
iswitch to Nsi, which, in turn, implies that in
the next generation, SD-OMN will attain convergence.
Case 3: If both Cases (1) and (2) occur, SD-OMN will
attain convergence. Hence, the proof.
Theorems 1 and 2 bear significance in the character-
izability of the convergence time of strategies in the
concerned SD-OMN. In other words, for given Kand
δ, one can determine if and when the OMN converges
to an equilibrium strategy. This is revisited in Section 7
in the context of experimental results.
Theorem 3: The message delivery ratio of the SD-OMN
increases with the increasing number of cooperators.
Accepted Version (For Personal Use Only)
8
TABLE 2: Datasets
Accepted Version (For Personal Use Only)
Dataset Mobility type Device Network type No. of devices Duration (days)
Infocom06 [25] Real iMote Bluetooth 78 4
PMTR2[26] Real PMTR RTX-RTLP 44 19
Sassy [27] Real T-mote Bluetooth 27 79
KAIST [28] Real T-mote GPS receiver 4 (92-trace) 78
HCM [29] Map-based - Bluetooth 100 5
RWP [29] Random way point - Bluetooth 100 5
Mathematically,
lim|NC|→|N|X
s∈S
αs(t)=1
Proof: Let |NC| |Ns0|, where s0∈S− {C}. The
chances of message delivery for a group gincreases
with the increase in the corresponding Ψg. So, we can
approximate αsas αs(t)≈Ψs
Ps∈SΨs. As the number of
cooperators is high, ΨCis also high, since the coopera-
tors forward the messages of exploiters as well. So, we
can write ΨCΨs0. Therefore,
αC(t) = ΨC
Ps∈SΨs
≈ΨC
ΨC
= 1.
Moreover, as Ψs0→0,αs0→0,s0∈S− {C}. Therefore,
X
s∈S
αs(t) = αC(t) + X
s0∈S−{C}
αs0(t)=1.
Hence, the proof.
Theorem 3 holds true in a SD-OMN, when one of the
following conditions is satisfied.
1) The traffic generation rate is low, but messages are
generated throughout the lifetime of the system
considered.
2) The traffic rate is high and messages are generated
for a short period of time (possibly during the
initial generation of the SD-OMN).
In the first case, due to low traffic rate, less number of
messages are generated in the time period considered.
Theoretically, almost all of them can be delivered to
the respective destination if most of the nodes are co-
operators, so that lim|NC|→|N|Ps∈Sαs(t)=1. In reality,
depending on the contact patterns among the nodes,
the same may be less than unity. In the second case,
although more messages are created in the system, a
comparatively longer lifetime allows the delivery of
most of them. Again, when most of the nodes are
cooperators, lim|NC|→|N|Ps∈Sαs(t)=1. Moreover, this
theorem verifies the goal function presented in Equation
2.
The case of high traffic rate, however, deserves further
discussion. In real life, there are practical constraints
on the buffer capacity of the nodes and/or lifetime
of the messages. Even if such constraints are relaxed,
continuous generation of messages throughout the life-
time of the system does not provide sufficient time to
the nodes to deliver them using the store-carry-and-
forward approach that is typical of OMNs. This can be
realized from the fact that in OMNs, the average message
delivery latency is typically a few thousand seconds,
or may be more. Therefore, in such cases, the claim of
Theorem 3 can be sub-optimal.
5.2 Complexity Analysis
Space complexity: Each node caches the CM ,DM, and
DP lists in its buffer. Let σ,ϕ,λ, and µbytes be
required to store a node’s address, message identity,
time-stamp, and delivery probability, respectively. Let
us further assume that a node generates Xnumber of
messages, out of which Ymessages are delivered. Let
Nbe the total number of nodes in the SD-OMN. We
compute the space required for storing Xmessages in
CM and Ymessages in DM. The space complexity for
storing CM ,DM, and DP at the nodes, respectively, are:
S(CM ) = X×(ϕ+λ)∈O(X), X (ϕ+λ)
S(DM ) = Y×(ϕ+λ)∈O(Y), Y (ϕ+λ)
S(DP ) = N×(σ+µ+λ)∈O(N), N σ+µ+λ)
So, the space complexity for storing these lists in a node
is O(n), where n=X+Y+N. As an illustration, let
σ= 8 byte (typical for Bluetooth hardware address),
ϕ=λ=µ= 4 byte and N= 100. So, the size of
the list DP is, |DP |= 100 ×(8 + 4 + 4) = 1600 byte.
Similarly, |C M|= 8Xbyte, and |DM |= 8Ybyte. So,
the total buffer overhead, B= 1600 + 8(X+Y)byte.
With X=Y= 100,B= 3200 byte. Figure 4 shows
the buffer requirement versus the number of messages
for storing the lists, when X=Y. This is insignificant,
given the fact that today’s smart phones have storage
capacities ranging upto a few gigabytes.
0
2
4
6
8
10
100 200 300 400 500
Buffer required (kilobyte)
Number of messages (X)
N = 100 N = 50 N = 25
Fig. 4: Storage overhead of DISCUSS as a function of the
number of messages.
9
Time complexity: The time complexity of Algorithm
1 is max{O(XY ), O(N2)}=O(m2), where m=N
or m=number of messages created or delivered. Thus,
the time complexity is quadratic in terms of the number
of nodes in the network, or the number of messages cre-
ated and delivered – whichever is maximum. The most
time consuming part of DISCUSS algorithm (as shown
in Figure 3) is the phase of acquiring information, which
is described in Algorithm 1. The remaining steps of the
DISCUSS algorithm are linear or constant, and do not
affect the order of complexity. So, the time complexity
of DISCUSS is O(m2).
5.3 DISCUSS with Global Knowledge
For the sake of completeness and evaluating the effec-
tiveness, we also consider a version of DISCUSS, where
the nodes have complete information about the SD-
OMN. In this case, we assume the presence of a Central
Authority (CA) in the network, with which the nodes can
communicate instantaneously. Whenever a new message
is created (or delivered), the CA is informed by the
concerned node. Based on these information, the CA
computes the DP of each node. At the end of each τ, all
the nodes get the DP information of all the nodes in the
network from the CA. Based on this, the nodes adapt
their strategies to the most successful one, if required.
6 SIMULATION DESIGN
This Section gives the overview of the simulation setup,
data sets, and metrics used for evaluation.
6.1 Simulation Setup
We evaluated the DISCUSS scheme using the ONE
simulator [29] considering the synthetic mobility traces,
Helsinki City Map (HCM) and Random Way Point (RWP),
together with the real-world traces, as summarized in
Table 2. For the HCM and RWP mobility models, we
executed the simulations for 2 days with 100 mobile
nodes deployed in terrains of size 4500 ×3400 m2and
450×340 m2, respectively, moving at a speed of 0.5−1.5
m/s. Messages were generated uniformly at random be-
tween every 30−40 seconds, with two randomly chosen
nodes as the source-destination pair. The message size
was chosen to be 25 KB and the Time-to-Live (TTL) of
the messages was 5hours. Each node had a buffer size of
3MB, transmission speed of 250 Kbps, and transmission
range of 10 meter.
The nodes used the binary Spray and Wait [3] routing
protocol with maximum 10 copies per message. How-
ever, it is important to point out that the proposed
approach is also applicable to other routing schemes.
At the beginning of the simulation, each node was
assigned to an initial strategy from the strategy set S, in
such a way that the percentage of the nodes having each
strategy was equal. In the static scenario, the nodes do
not change their respective strategies. On the other hand,
in the dynamic scenario, at the end of each generation
interval, the nodes decided whether or not to change
their respective strategies. Based on the information
available at the end of every generation interval (τ),
each node determined the most successful strategy and
switched to it if its strategy was not already the same.
Unless otherwise specified, the generation interval was
taken to be 1 hour.
The average value of the results taken over 30 runs to-
gether with the corresponding 95% confidence intervals
are reported.
6.2 Performance Metrics
We evaluated the performance of the proposed scheme,
DISCUSS, using the following metrics.
6.2.1 Group composition (size)
The percentage of cooperators, exploiters, and isolators
in each generation with respect to time.
6.2.2 Delivery ratio of messages
It is the ratio of the total number of messages success-
fully delivered to the intended destination nodes to the
total number of messages generated in the OMN.
6.2.3 Average message delivery latency
It is the average time taken for delivering a message from
the source node to its corresponding destination node.
We also compared the results of performance eval-
uation obtained for three variants of the DISCUSS al-
gorithm: (a) DISCUSS in a dynamic scenario (referred
to as “dynamic” in the plots), (b) DISCUSS in a static
scenario (referred to as “static” in the plots), and (c)
DISCUSS with global knowledge, as discussed in Section
5.3 (referred to as “global” in the plots).
7 SIMULATION RESULTS
The results obtained in this manuscript are derived
considering the datasets given in Table 2. In this Section,
we present some results of DISCUSS using the metrics
given in Section 6.2.
7.1 Effects of Generation Interval
Figure 5 shows the delivery ratio of messages for dif-
ferent mobility scenarios and different generation in-
tervals (τ). In the KAIST and RWP mobility scenarios,
the message delivery ratio was higher when τ= 1
hour, and it decreased as τincreased. This is because,
as τincreased, the number of exploiters and isolators
remained the same in respective strategies for τtime.
So, the message drops increased in the OMN. It was
observed that in 1hour, almost all nodes had information
about the other nodes. So, they attempted to switch
their respective strategies. The sooner they changed their
strategies, the better they earned in terms of delivery
ratio. So, the overall network-wide message delivery
Accepted Version (For Personal Use Only)
10
0.2
0.3
0.4
0.5
0.6
1 2 3 4 5 6 7 8 9 10 11 12
Delivery ratio
Generation interval (hour)
(a) Infocom06
0.08
0.09
0.1
0.11
0.12
0.13
1 2 3 4 5 6 7 8 9 10 11 12
Delivery ratio
Generation interval (hour)
(b) Sassy
0.08
0.1
0.12
0.14
0.16
1 2 3 4 5 6 7 8 9 10 11 12
Delivery ratio
Generation interval (hour)
(c) KAIST
0.3
0.4
0.5
0.6
0.7
1 2 3 4 5 6 7 8 9 10 11 12
Delivery ratio
Generation interval (hour)
(d) HCM
0.3
0.4
0.5
0.6
0.7
1 2 3 4 5 6 7 8 9 10 11 12
Delivery ratio
Generation interval (hour)
(e) RWP
0.24
0.25
0.26
0.27
0.28
1 2 3 4 5 6 7 8 9 10 11 12
Delivery ratio
Generation interval (hour)
(f) PMTR
Fig. 5: Message delivery ratio for different generation intervals.
increased, as the generation interval was less. Similarly,
using the Infocom06 trace as well, the message delivery
ratio increased when τ= 1 hour. But in the HCM
scenario, the delivery ratio improved marginally, when
τ= 2 hour, because in this case, in the 1st hour, the
number of direct contacts was more than the indirect
contacts, which implies that almost all messages of nodes
are delivered. So, less number of nodes changed their
strategies. This is ascribed to the fact that delivery ratio
was high when τ≈1hour. So, hereafter, we consider
τ= 1 hour for all simulations. We present the exper-
imental results using the map-based (HCM), random
way point (RWP) mobility model and real mobility traces
(Infocom06 [25], KAIST [28], Sassy [27], and PMTR [26]).
7.2 Similarity Measurement
Similarity index is used to determine the similarity be-
tween two documents. Using this metric, we established
the effectiveness of DISCUSS by comparing it with its
variant, in which global knowledge is available. Jac-
card similarity index has been widely used in several
works related to DTNs such as in the work by Hui
and Crowcroft [30] as well as, wireless networks [31],
in general. The application of Jaccard similarity index
is suitable to evaluate the similarity of two sets, as
considered here, unlike many other similarity indices
such as, cosine similarity index, which operates upon
two vectors. Let, after a generation interval, NDand
NG, respectively, be the set of nodes using DISCUSS
and its variant with global knowledge. In particular,
|NG|=|ND|=|N|;NGand NDdiffer only in terms of
the strategies adapted by the different nodes, e.g., node
imay have strategy Cin NG, and Ein ND. Then, the
Jaccard similarity index [18] is defined as:
Js=|NG∩ND|
Accepted Version (For Personal Use Only)
|NG∪ND|(13)
A measure of Js∈[0,1] indicates the closeness in per-
formance of DISCUSS with local and global knowledge.
0
0.2
0.4
0.6
0.8
1
Infocom06 HCM KAIST Sassy PMTR RWP
Similarity
1h 5h 10h 15h 20h 1d 2d
Fig. 7: Jaccard Similarity Index between DISCUSS and
its variant with global knowledge.
Figure 7 shows the similarities between DISCUSS with
global and local knowledge for different time intervals.
It was observed that the value of Jaccard similarity
index increased with time, since the nodes obtained
more accurate information as time increased. For the
Infocom06, HCM, KAIST, and RWP data sets, the Jaccard
similarity index was observed to be above 85% at the end
of 2 days. However, for Sassy and PMTR, the similarities
varied between 55% to 70%. In RWP, initially, in the 1st
hour, very less number of nodes changed their strategies.
So, the similarity is higher at the beginning. In RWP,
the nodes met less frequently and for short durations.
Therefore, using the DISCUSS scheme, the nodes did
not exhibit accurate possessing information about other
groups. So, the similarity decreased significantly (30%)
between 1 to 10 hours using RWP.
7.3 Effects on the Group Composition
Figure 6 shows the variation in group composition with
time. Initially, in all the simulation scenarios, the per-
11
0
20
40
60
80
100
0 1h 4h 8h 12h 1d 2d
Group size (%)
Time
Cooperators Exploiters Isolators
(a) HCM
0
20
40
60
80
100
0 1h 4h 8h 12h 1d 2d
Group size (%)
Time
Cooperators Exploiters Isolators
(b) RWP
0
20
40
60
80
100
0 1h 4h 8h 12h 1d 2d
Group size (%)
Time
Cooperators Exploiters Isolators
(c) KAIST
0
20
40
60
80
100
0 1h 4h 8h 12h 1d 2d
Group size (%)
Time
Cooperators Exploiters Isolators
(d) Infocom06
0
20
40
60
80
0 1h 4h 8h 12h 1d 2d
Group size (%)
Time
Cooperators Exploiters Isolators
(e) PMTR
0
20
40
60
80
100
0 1h 4h 8h 12h 1d 2d
Group size (%)
Time
Cooperators Exploiters Isolators
(f) Sassy
Fig. 6: Variation in group composition with time.
centages of cooperators, exploiters, and isolators were
the same. In all the scenarios, the residual number of
cooperators varied between 85-100% and the number of
exploiters and isolators varied between 0-10%, except for
the PMTR case. Using the PMTR trace, the percentage
of cooperators, exploiters, and isolators were 78%, 18%,
and 4%, respectively. We observed from Figure 6 that
the group size of the cooperators increased with time.
The reason behind this is that, generally, the weighted
delivery ratio (Equation (10)) of the cooperators is more,
because the cooperators help in forwarding other nodes’
messages. So, after each generation interval, more nodes
switched their strategy to cooperate, instead of exploit
and isolate. Another interesting phenomenon that was
observed is that for all the traces, approximately after
1 day, 80% nodes chose cooperation as their strategy
because cooperate is the most successful strategy. In the
HCM, RWP, KAIST, and Infocom06 traces, the number
of isolators reached 0% after 1 day and number of
exploiters reached 2-3%, except in the case of the HCM
and PMTR traces.
(K, δ)convergence: From Figure 6(d), it can be observed
that beyond 24 hour, the percentage of cooperators is at
least 91%. Let us consider the time duration between 24
hour and 36 hour consisting of 12 generations, with τ= 1
hour. We can infer from Figure 6(d) that the SD-OMN
attained (K, δ)convergence, with K= 12 and δ= 0.1.
Similar inferences can be drawn from others as well.
7.4 Delivery Ratio of Messages
We evaluated DISCUSS with respect to the static,dy-
namic, and global scenarios. Figure 8 shows the cumu-
lative delivery ratio as a function of time. Here, first
we discuss the comparison of DISCUSS in the static and
dynamic scenarios, and then we discuss DISCUSS in the
global and dynamic scenarios.
It can be observed that in all the mobility traces,
DISCUSS, in the dynamic scenario exhibited higher de-
livery ratio, than in the static scenario. For example, in
the Infocom06 and HCM traces, the dynamic scenario
outperformed the static one by 31% and 17%, respec-
tively, at the end of 2 days. The reason behind this is
that, in the static scenario, the group-wise percentage
of nodes remained the same throughout the simulation.
As discussed earlier, exploiters and isolators do not
cooperate in forwarding other nodes’ messages. So, the
message drop probability increased over time and the
delivery probability decreased for the static scenario. On
the other hand, in the dynamic scenario, the group-wise
percentage of nodes changed over time and most of the
nodes switched their strategy to cooperate. Hence, the
message delivery ratio increased over time.
Further, in the Infocom06 trace, the message delivery
ratio was less between 2-12 hours, because during this
time, the number of contacts per hour was quite less, as
shown in Figure 9(b). So, during this time, less number
of messages were delivered to the destination nodes.
However, in the HCM mobility trace, as shown in Figure
9(a), the number of contacts varied between 350 to 550,
which is nearly similar throughout the simulation dura-
tion. So, the probability of message delivery increased
over time when the cooperators increased with time.
In KAIST mobility trace, the message delivery ratio
decreased after 12 hour, because the number of con-
tacts was observed to be less among nodes, in between
12 −48 hour and message generation rate was uniform
Accepted Version (For Personal Use Only)
12
0
0.2
0.4
0.6
0.8
1
1h 4h 8h 12h 1d 2d
Delivery ratio
Time
Dynamic
Static
Global
(a) HCM
0
0.2
0.4
0.6
0.8
1
1h 4h 8h 12h 1d 2d
Delivery ratio
Time
Dynamic
Static
Global
(b) RWP
0
0.2
0.4
0.6
0.8
1
1h 4h 8h 12h 1d 2d
Delivery ratio
Time
Dynamic
Static
Global
(c) KAIST
0
0.2
0.4
0.6
0.8
1
1h 4h 8h 12h 1d 2d
Delivery ratio
Time
Dynamic
Static
Global
(d) Infocom06
0
0.2
0.4
0.6
0.8
1
1h 4h 8h 12h 1d 2d
Delivery ratio
Time
Dynamic
Static
Global
(e) PMTR
0
0.2
0.4
0.6
0.8
1
1h 4h 8h 12h 1d 2d
Delivery ratio
Time
Dynamic
Static
Global
(f) Sassy
Fig. 8: Message delivery ratio using DISCUSS in different scenarios.
Accepted Version (For Personal Use Only)
100
200
300
400
500
600
1 6 12 18 24 30 36 42 48
Number of contacts
Time (hour)
(a) HCM
0
2000
4000
6000
8000
10000
1 6 12 18 24 30 36 42 48
Number of contacts
Time (hour)
(b) Infocom06
Fig. 9: Number of contacts in the network.
throughout the simulation duration. The delivery ratio
of RWP, KAIST, PMTR, and Sassy were 51%, 15%, 25%,
and 16% respectively, due to assorted number of contacts
on different mobility scenarios.
We compared the delivery ratio of the messages in
global and dynamic scenarios. For the global scenario,
accurate and same information were possessed by all
the nodes and each node took decision accordingly. This
implies that the delivery ratio for the global case is
more accurate than the distributed case in the dynamic
scenario. In the KAIST, PMTR, and Sassy scenarios, the
delivery ratio was almost similar for both the global and
dynamic scenarios, as the number of contacts between
nodes were higher and repetitive. In RWP, the delivery
ratio of DISCUSS, in the global scenario was more than
that in the dynamic scenario, because in the later case,
the nodes did not get the complete information about the
other groups of nodes due to less number of contacts.
However, in the HCM scenario, the contacts were more
and repetitive. Initially, the nodes had partial informa-
tion about network performance. As time progressed,
they came in contact with more number of nodes. So,
the delivery ratio in the case of the dynamic scenario
was almost the same as that of the global scenario after
1 day. Similarly, in the Infocom06 scenario, the delivery
probability of the dynamic scenario was approximately
the same as that in the global scenario.
0.4
0.5
0.6
0.7
0.8
0.9
1
20 40 60 80 100
Delivery ratio
Cooperators (%)
20-30
30-40
40-50
50-60
(a) HCM
0.3
0.4
0.5
0.6
0.7
20 40 60 80 100
Delivery ratio
Cooperators (%)
20-30
30-40
40-50
50-60
(b) Infocom06
Fig. 11: Variation in percentage of cooperators under
light traffic load.
Figure 11 shows the delivery ratio of the messages
in SD-OMN, under light traffic load, which validates
the claim of Theorem 3. Here, the message generation
duration was taken as 1day. We considered the SD-
OMN with static scenarios, by varying the percentage of
cooperators and message generation rates. It is observed
from Figure 11 that as the percentage of cooperators
increased, the delivery ratio of the messages increased,
correspondingly, towards unity.
7.5 Message Delivery Latency
Figure 10 shows the average delivery delay of the mes-
sages of the three scenarios on different mobility data
sets. The average delivery latency of DISCUSS in the
dynamic scenario was observed to be almost similar
to that in the global scenario, for all the mobility data
13
0
1500
3000
4500
6000
7500
9000
1h 4h 8h 12h 1d 2d
Average delivery latency (second)
Time
Dynamic
Static
Global
(a) HCM
0
1000
2000
3000
4000
5000
6000
7000
1h 4h 8h 12h 1d 2d
Average delivery latency (second)
Time
Dynamic
Static
Global
(b) KAIST
0
1500
3000
4500
6000
7500
9000
1h 4h 8h 12h 1d 2d
Average delivery latency (second)
Time
Dynamic
Static
Global
(c) RWP
Accepted Version (For Personal Use Only)
0
1500
3000
4500
6000
7500
9000
1h 4h 8h 12h 1d 2d
Average delivery latency (second)
Time
Dynamic
Static
Global
(d) Infocom06
0
5000
10000
15000
20000
25000
1h 4h 8h 12h 1d 2d
Average delivery latency (second)
Time
Dynamic
Static
Global
(e) PMTR
0
2000
4000
6000
8000
10000
12000
1h 4h 8h 12h 1d 2d
Average delivery latency (second)
Time
Dynamic
Static
Global
(f) Sassy
Fig. 10: Message delivery delay using DISCUSS in different scenarios.
sets, except the Sassy mobility trace. Figure 10 shows
some variation in delay between the global and dynamic
cases because of incomplete information. In the HCM,
Infocom06, and RWP cases the average delivery latency
of the messages in the dynamic scenario is less than that
in the static scenario. In these mobility data sets, the
number of inter-node contacts is more. So, as the size of
the cooperator increased, more messages were delivered
and the delivery delay was reduced in the dynamic
scenario. For example, using the HCM mobility trace,
latency decreased to 136% at the end of 2 days. In the
Infocom06 trace, initially the delay was more between
4 to 12 hour, even in the presence of more number of
cooperators, as less number of contacts are available. As
the number of contacts increased, the delay gradually
decreased. In the KAIST, PMTR, and Sassy traces, the
average delay in the dynamic scenario is more than that
in the static scenario, because less number of contacts
are available. So, the message delivery latency is less.
8 CONCLUSION
In this paper, we proposed a distributed cooperation
mechanism, named DISCUSS, for stimulating the non-
cooperative nodes to cooperate for forwarding the other
nodes’ messages. After a generation interval, each node
compares its own performance with the performance of
the other groups and selects the most successful strategy
accordingly. We verified the similarity and effectiveness
of DISCUSS in different scenarios. We compared the per-
formance of dynamic DISCUSS with the static DISCUSS,
without changing the initial chosen strategies. Extensive
simulation results show that the dynamic DISCUSS is
better than the static DISCUSS. In the future, this work
can be extended by exploring other possibilities of non-
cooperative node strategies. Also, the proposed mecha-
nism can be extended by considering the privacy and
security aspects of SD-OMN. We, further, plan in the
future to validate the results using a real-life test-bed.
ACKNOWLEDGEMENT
This work was partially supported by a fellowship spon-
sored by the Tata Consultancy Services (TCS), India.
REFERENCES
[1] K. R. Fall, “A delay-tolerant network architecture for challenged
internets,” in Proc. of the SIGCOMM, Germany, 2003, pp. 27–34.
[2] A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass, and J. Scott,
“Impact of human mobility on opportunistic forwarding algo-
rithms,” IEEE Transaction on Mobile Computing, vol. 6, no. 6, pp.
606–620, 2007.
[3] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Spray and
wait: an efficient routing scheme for intermittently connected
mobile networks,” in Proceedings of ACM SIGCOMM Workshop on
Delay-tolerant Networking, New York, USA, 2005, pp. 252–259.
[4] A. Vahdat and D. Becker, “Epidemic routing for partially-
connected ad hoc networks,” Duke University, Tech. Rep. CS-
2000-06, 2000.
[5] A. Lindgren, A. Doria, and O. Schelen, “Probabilistic routing in
intermittently connected networks,” SIGMOBILE Mobile Comput-
ing Communication Review, vol. 7, pp. 19–20, 2003.
[6] Q. Yuan, I. Cardei, and J. Wu, “An efficient prediction-based
routing in disruption-tolerant networks,” IEEE Transactions on
Parallel and Distributed Systems, vol. 23, no. 1, pp. 19–31, 2012.
[7] T. Abdelkader, K. Naik, A. Nayak, N. Goel, and V. Srivastava,
“SGBR: A routing protocol for delay tolerant networks using
social grouping,” IEEE Transactions on Parallel and Distributed
Systems, vol. 24, no. 12, pp. 2472–2481, 2013.
[8] H. Zhu, S. Du, Z. Gao, M. Dong, and Z. Cao, “A probabilistic
misbehavior detection scheme toward efficient trust establishment
in delay-tolerant networks,” IEEE Transactions on Parallel and
Distributed Systems, vol. 25, no. 1, pp. 22–32, 2014.
Accepted Version (For Personal Use Only)
14
[9] H. Zhu, X. Lin, R. Lu, Y. Fan, and X. S. Shen, “Smart: A
secure multilayer credit-based incentive scheme for delay-tolerant
networks,” IEEE Transactions on Vehicular Technology, vol. 58, pp.
4628–4639, 2009.
[10] U. Shevade, H. H. Song, L. Qiu, and Y. Zhang, “Incentive-aware
routing in dtns,” in Proceedings of the 16th IEEE International
Conference on Network Protocols, Orlando, USA, 2008, pp. 238–247.
[11] B. B. Chen and M. C. Chan, “MobiCent: a credit-based incentive
system for disruption tolerant network,” in Proceedings of INFO-
COM, San Diego, California, USA, 2010, pp. 875–883.
[12] L. Wei, H. Zhu, Z. Cao, and X. Shen, “MobiID: a user-centric
and social-aware reputation based incentive scheme for de-
lay/disruption tolerant networks,” in Proceedings of the 10th
International conference on Ad-hoc, Mobile, and Wireless Networks,
Germany, 2011, pp. 177–190.
[13] X. Zhang, X. Wang, A. Liu, Q. Zhang, and C. Tang, “Pri: A
practical reputation-based incentive scheme for delay tolerant
networks,” KSII Transactions on Internet and Information Systems,
vol. 6, no. 4, pp. 973–988, 2012.
[14] F. Wu, T. Chen, S. Zhong, C. Qiao, and G. Chen, “A game-theoretic
approach to stimulate cooperation for probabilistic routing in
opportunistic networks,” IEEE Transactions on Wireless Communi-
cations, vol. 12, no. 4, pp. 1573–1583, 2013.
[15] R. Lu, X. Lin, H. Zhu, X. Shen, and B. Preiss, “Pi: A practical
incentive protocol for delay tolerant networks,” IEEE Transactions
on Wireless Communications, vol. 9, no. 4, pp. 1483–1493, 2010.
[16] S. Pal, S. Misra, and B. K. Saha, “Rock-scissors-paper cycle of
cooperation strategies in opportunistic mobile networks,” in 3rd
IEEE International Workshop on Smart Communication Protocols and
Algorithms (SCPA 2013), Hungary, 2013, pp. 1062–1066.
[17] R. B. Freeman, The works of Charles Darwin: an annotated biblio-
graphical handlist,2nd ed., F. Darwin, Ed. Dawsons, 1977.
[18] P. Jaccard, “Etude comparative de la distribution florale dans une
portion des alpes et des jura,” Bulletin de la Societe Vaudoise des
Sciences Naturelles, vol. 37, pp. 547–579, 1901.
[19] S. Misra, S. Pal, and B. K. Saha, Next-Generation Wireless Technolo-
gies. Springer, 2013, ch. Cooperation in Delay Tolerant Networks,
pp. 15–35.
[20] G. Resta and P. Santi, “A framework for routing performance
analysis in delay tolerant networks with application to non co-
operative networks,” IEEE Transactions on Parallel and Distributed
Systems, vol. 23, no. 1, pp. 2–10, 2012.
[21] L. Buttyan, L. Dora, M. Felegyhazi, and I. Vajda, “Barter trade
improves message delivery in opportunistic networks,” Ad Hoc
Networks, vol. 8, pp. 1–14, 2010.
[22] L. Yin, H. M. Lu, Y. D. Cao, and J. M. Gao, “Cooperation in
delay tolerant networks,” in 2nd International Conference on Signal
Processing Systems, Dalian, China, 2010, pp. V1–202–V1–205.
[23] A. Elwhishi, P. H. Ho, K. Naik, and B. Shihada, “A novel message
scheduling framework for delay tolerant networks routing,” IEEE
TPDS, vol. 24, no. 5, pp. 871–880, 2013.
[24] D. J. Futuyma, Evolution,2nd ed. Massachusetts, USA: Sinauer
Associates, 2009.
[25] J. Scott, R. Gass, J. Crowcroft, P. Hui, C. Diot, and A. Chaintreau,
“CRAWDAD trace (v. 2009-05-29),” Downloaded from
http://crawdad.cs.dartmouth.edu/cambridge/haggle/imote/infocom2006,
May 2009.
[26] P. Meroni, S. Gaito, E. Pagani, and G. P. Rossi, “CRAW-
DAD data set unimi/pmtr (v. 2008-12-01),” Downloaded from
http://crawdad.cs.dartmouth.edu/unimi/pmtr, Dec 2008.
[27] G. Bigwood, D. Rehunathan, M. Bateman, T. Hen-
derson, and S. Bhatti, “CRAWDAD data set stan-
drews/sassy (v. 2011-06-03),” Downloaded from
http://crawdad.cs.dartmouth.edu/standrews/sassy, Jun 2011.
[28] I. Rhee, M. Shin, S. Hong, K. Lee, S. Kim, and S. Chong, “CRAW-
DAD trace ncsu/mobilitymodels/gps/kaist,” Downloaded from
http://crawdad.cs.dartmouth.edu/ncsu/, July 2009.
[29] A. Ker¨
anen, J. Ott, and T. K¨
arkk¨
ainen, “The ONE simulator for
DTN protocol evaluation,” in Proceedings of the 2nd International
Conference on SIMUTools, Belgium, 2009, pp. 55:1–55:10.
[30] P. Hui and J. Crowcroft, “Predictability of human mobility and
its impact on forwarding,” in In 3rd International Conference on
Communications and Networking, China, 2008, pp. 543–547.
[31] J. M. Bahi, A. Makhoul, and M. Medlej, “Data aggregation for
periodic sensor networks using sets similarity functions,” in
Proceedings of the 7th International Conference on Wireless Commu-
nications and Mobile Computing, Turkey, 2011, pp. 559–564.
Dr. Sudip Misra is an Associate Professor in the
School of Information Technology at the Indian
Institute of Technology Kharagpur. He received
his Ph.D. degree in Computer Science from Car-
leton University, Ottawa, Canada. His research
interest is broadly on wireless networks. He has
won 8 research paper awards in different confer-
ences. He was awarded the IEEE ComSoc Asia
Pacific Outstanding Young Researcher Award at
IEEE GLOBECOM 2012. He was also awarded
the Canadian Governments NSERC Post Doc-
toral Fellowship and the Humboldt Research Fellowship in Germany.
Sujata Pal is currently pursuing her PhD degree
from the School of Information Technology at the
Indian Institute of Technology Kharagpur, India.
She received the MTech degree in Multimedia
and Software System from the West Bengal
University of Technology, India, in 2007 and the
BE degree in Computer Science from the North
Orissa University, India, in 2002. She has more
than 5 years of teaching experience. Her current
research interests are DTNs and MANETs.
Barun Kumar Saha is pursuing MS at School of
Information Technology, Indian Institute of Tech-
nology Kharagpur. His research interests include
MANETs, DTNs, OMNs together with human
aspects and their applications, and the use of
technology for advancing education. He is the
lead developer of the Software Engineering and
Advanced Network Technologies Virtual Labs.
He maintains a tutorial on DTNs at http://delay-
tolerant-networks.blogspot.com/