Content uploaded by Tamer Mostafa Abdelkader
Author content
All content in this area was uploaded by Tamer Mostafa Abdelkader on Apr 21, 2018
Content may be subject to copyright.
1
A Performance Comparison of Delay Tolerant
Network Routing Protocols
Tamer Abdelkader, Kshirasagar Naik, Amiya Nayak, Nishith Goel, and Vineet Srivastava
Abstract—Networks that lack continuous end-to-end connec-
tions among their nodes due to node mobility, constrained power
sources, or limited data storage space are called Delay Tolerant
Networks (DTN). To overcome the intermittent connectivity, DTN
nodes store and carry the data packets they receive until they
come into communication range of each other. In addition,
they spread multiple copies of the same packet on the network
to increase the delivery probability. In recent years, several
routing protocols have been developed specifically for DTN.
These protocols vary in the number of copies they spread
and the information they use to guide the packets to their
destinations. There have been some reviews of those protocols,
but no performance comparison has been conducted.
In this paper, we study four well-known DTN routing pro-
tocols: EPIDEMIC, Spray-and-Wait (SnW), PROPHET and
MAXPROP. We introduce a procedural form to present the
protocols. We measure the performance of the protocols in
terms of packet delivery, delivery cost, and average packet
delay. We compare the protocols performance together with the
results of optimal routing using real-life scenarios of vehicles and
pedestrians roaming in a city. We conduct several simulation
experiments to show the impact of changing buffer capacity,
packet lifetime, packet generation rate, and number of nodes on
the performance metrics. The paper is concluded by providing
guidelines to develop an efficient DTN routing protocol. To the
best of our knowledge, this work is the first to provide detailed
performance comparison among the diverse collection of DTN
routing protocols.
Index Terms—Delay Tolerant Networks, Routing, Performance
comparison.
I. INTRODUCTION
WITH the growing scope of applications of handheld
devices, such as smartphones and tablet computers, and
the need to access the Internet anywhere and anytime, wide-
area mobile ad-hoc networks have been deployed spanning
entire cities. Such networks involve pedestrians and vehicles
on the roads, existing infrastructure on the road sides, and
even devices inside buildings. These types of heterogeneous
networks are characterized by their fast and continuously
changing topologies, a diverse variety of resource-constraints
such as buffer, energy, and bandwidth, and the frequent
Tamer Abdelkader is with the Dept. of Information systems, Faculty of
Computers and Information Sciences, Ain Shams University, Cairo, Egypt.
Email: tammabde@cis.asu.edu.eg.
Kshirasagar Naik is with the Dept. of Electrical & Computer Engineer-
ing, University of Waterloo, Waterloo, Ontario, Canada, N2L3G1. Email:
snaik@uwaterloo.ca.
Amiya Nayak is with the School of Information Technology & Engi-
neering, University of Ottawa, Ottawa, Ontario, Canada, K1N 6N5. Email:
anayak@site.uottawa.ca.
Nishith Goel and Vineet Srivastava are with Cistech Limited, 210
Colonnade Road, Unit 3, Ottawa, Ontario, Canada, K2E 7L5. Email:
ngoel@cistech.ca, vineet@cistel.com.
partitioning among its nodes. Because of these challenges,
data is handled on a best-effort basis, and the objective
becomes the delivery of as much data as possible with the
minimum resource consumption. These networks are called
Delay Tolerant Networks (DTN) [1].
Applications of DTN have been found in many challeng-
ing environments such as providing delay-tolerant Internet
services in suburban and rural areas. This project has been
implemented by First Mile Solutions with a system called
DakNet [2]. Vehicular networking is a wide and growing field
of DTNs, where many applications are being explored [3].
One of these applications is to provide Internet access to
vehicles, by connecting to roadside wireless base stations [4].
Non-commercial applications include monitoring and tracking
wildlife animals [5], and environmental monitoring, such as
lake water quality monitoring and road-side noise monitoring
[6]. DTNs can be applied in a variety of other fields ranging
from healthcare to education to economic efficiency.
Traditional routing protocols for wired and wireless net-
works fail to work in this environment because they assume
the existence of continuous end-to-end connections between
sources and destinations. Routing protocols developed for
DTN adapt themselves to this challenging environment by
propagating multiple copies of data packets to increase the
probability that one of the copies reaches the destination.
Nodes receiving the packet copies store them until they meet
other nodes or meet their destinations. Simple DTN routing
protocols blindly forward packet copies to any node they
become in contact without using a selection criteria. They
range from the full network flooding to the partial flooding.
The Blind-Routing approach may achieve high delivery ratio
of data if provided enough storage and energy resources. On
the other hand, it has its drawbacks, such as burdening the
buffer and the inefficient use of the contact duration. Other
routing protocols tend to restrict forwarding of data packets
to selected nodes. Using some information collected from the
network, they guide the packet copies to their destinations by
selecting the relay nodes. The Guided-Routing approach fails
if the network topology is changing faster than the rate of
information gathering.
From the analysis of current DTN routing protocols, it was
found that there are several trade-offs to be considered in a
protocol design:
•A trade off between maximizing packet delivery ratio
and minimizing the number of transmissions in the net-
work. Maximizing delivery ratio requires increasing the
number of packet copies spread throughout the network
to increase the probability of reaching the destination,
2
while minimizing the number of transmissions reduces
energy consumption and network overhead, but decreases
the number of packet copies.
•Another trade off is between collecting information to
use it in selecting forwarding nodes or not collecting
network information and randomly forward the packets.
Collecting information from the network helps in finding
better routes to destinations, but requires time to collect
the information which increases packet delays. On the
other hand, not collecting information leads to spreading
the packet copies blindly, and decreases the probability of
reaching the destination unless a large number of copies
were spread.
In this paper, we conduct simulations using real life scenar-
ios of pedestrians and vehicles roaming in a city. Our goal is
to compare the performance of four well-known DTN routing
protocols:
•EPIDEMIC [7]
•Spray-and-Wait (SnW) [8]
•PROPHET [9], and
•MAXPROP [10]
While there are many routing protocols developed in the
recent years, we selected these four protocols because they
are leading in their distinguishing behavior, and they are well
referenced in the other protocols. EPIDEMIC is an example
of a Blind-Routing Full-Flooding protocol, and is historically
the first protocol developed for DTN. SnW is an example of
a Blind-Routing Partial-Flooding protocol. PROPHET is an
example of a Guided-Routing protocol with a first in first
out (FIFO) packet selection mechanism. MAXPROP is an
example of a Guided-Routing protocol which favors packets
with minimum number of hops. A detailed explanation of these
protocols is provided in Section II. We compare the routing
protocols with each other and with the performance results
of optimal routing presented in [11]. We study the impact of
changing:
•Buffer Capacity,
•packet lifetime or time-to-live (TTL),
•node density,
•traffic load, and
•mobility model by using real-life simulations, while
model-based simulations have been used in [12].
The performance metrics used to compare the protocols are:
Delivery ratio, delivery cost and average packet delay. The
metrics are explained in Section III.
Our contributions can be summarized as follows:
•A procedural presentation of the routing protocols is
introduced together with a summary of similarities and
differences.
•A performance comparison among the routing protocols
and the optimal results is conducted in terms of the
performance metrics mentioned before.
•Analysis of the results and conclusions are drawn to
identify the main design issues to be considered in
designing a DTN routing protocol.
This work is intended to help network designers and managers
in selecting and developing routing protocol that fits delay
tolerant applications. A summary of these guidelines is given
in Section V.
The rest of the paper is organized as follows. In Section
II, we review the DTN routing protocols and present them in
a procedural form. The performance metrics and simulation
setup are presented in Section III. Simulation results are dis-
cussed in Section IV. The key design issues of a DTN routing
protocol are identified in Section V. Finally, conclusions are
drawn in Section VI.
II. DTN ROUT IN G PROTO CO LS
Routing protocols are classified according to the amount
and type of information used to take the routing decision [13].
Blind-routing protocols [7], [8] aim at fast spreading of packets
in the network. These protocols do not collect information
about other nodes because they do not use a node selection
criteria. They vary according to their spreading mechanism and
amount. Guided-Routing protocols [9], [10], [12], [14] aim at
efficiently selecting the relay nodes to enhance the delivery
probability in case of limited storage and energy resources.
To select relay nodes, they have to collect information about
other nodes in the network. Guided-Routing Protocols vary in
the type and amount of information gathered.
A. Epidemic Routing Protocol
Epidemic routing [7] is the first routing protocol proposed
for sparse networks. Each packet generated is assigned a
unique ID that is associated with it and all its copies till they
are dropped or delivered to the destination. The list of all the
packets IDs in a node’s buffer is called the summary vector.
When two nodes meet, they exchange their summary vector.
All data packets that are stored in one node and not in the
other are ordered on a first come first serve (FCFS) basis to
be transmitted to the other node. Packet transfer then starts
until the contact duration ends. Assuming that the contact
duration is long enough to transfer all the uncommon packets,
then the two nodes will have the same packet list after their
contact ends. Given unlimited buffer size, long enough contact
durations, unlimited lifetime for the data packets, and non-
infinite partitioned network, Epidemic routing guarantees the
delivery of all the packets to their destinations. In addition, it
guarantees the lowest end-to-end delay, because each packet
is routed on all possible paths from the source, and one of
the copies will be on the shortest path. Procedure 1 shows a
pseudo-code of the Epidemic protocol during the contact of
two nodes.
The main drawback of Epidemic routing is its huge con-
sumption of the limited resources, such as memory, energy and
contact duration. Later work have been proposed to reduce the
inefficient resource consumption.
B. Spray and Wait (SnW) Routing Protocol
DTN usually involve devices that are energy-sensitive in
which saving energy becomes one of its main objectives, if not
the main one. Energy consumption is mostly incurred in the
communication process (transmission and reception). To save
3
Algorithm 1 Epidemic Routing Protocol
1: Procedure Name: OnContact
2: Input: node a, node b, integer ContactDuration
3: DropExpiredPackets(a,b) /* Drop packets with their life-
time expired in both nodes */
4: ExchangeSummaryVector(a,b)
5: if ContactDuration > 0then
6: pkt=GetPacket(a)
7: if pkt then
8: if NotReceivedBefore(pkt,b)then
9: if IsDestination(pkt,b)then
10: SendPacket(pkt,a)
11: ConsumePacket(pkt,b)
12: else
13: SendPacket(pkt,a)
14: StorePacket(pkt,b)
15: end if
16: ContactDuration=ContactDuration-size(pkt)
17: end if
18: end if
19: end if
energy, it is required to decrease the number of transmissions
and receptions. Motivated by this fact, the authors in [8]
proposed the Spray-and-Wait (SnW) routing protocol. The
idea of SnW is to limit the number of packet copies in the
network. A packet copy, transferred from a node to another,
is associated with the number of further copies allowed for
the second node to distribute. This number is decreased by
the number of transfers for this packet at each node. When
the allowed number of copies reaches one, the carrying node
stops generating any more copies of the packet and keep its
single copy until it either meets the destination or the packet
is dropped because of a buffer overflow or lifetime expiry. A
binary version of SnW is also proposed in [8], in which each
node is allowed to use half the number of copies allowed for
the packet and the other half is left for the receiving node.
The pseudo-code for the binary SnW is shown in Procedure
2.
Both versions of SnW, regular and binary, proved to perform
better than the full flooding protocol, Epidemic, in terms of
average packet delay and energy consumption. However, SnW
still suffers from the blind selection of the next-hop nodes
which may degrade the packet delivery ratio.
C. PROPHET Routing Protocol
The Probabilistic Routing Protocol using History of En-
counters and Transitivity (PROPHET) is proposed in [9]. The
protocol estimates a node metric called delivery predictability,
P(a, b), at each node afor each destination b. When two
nodes meet, they update their delivery predictability towards
each other. Then the two nodes exchange their delivery pre-
dictability list towards other nodes to each other to update
their delivery predictability towards the other nodes using the
following equations:
Algorithm 2 Binary Spray And Wait Routing Protocol
1: Procedure Name: OnContact
2: Input: node a, node b, integer ContactDuration
3: DropExpiredPackets(a,b) /* Drop packets with their life-
time expired in both nodes */
4: ExchangeSummaryVector(a,b)
5: if ContactDuration > 0then
6: pkt=GetPacket(a)
7: if pkt then
8: if NotReceivedBefore(pkt,b)then
9: if IsDestination(pkt,b)then
10: SendPacket(pkt,a)
11: ConsumePacket(pkt,b)
12: else
13: NrOf C opies=GetNrOfCopies(pkt,a)
14: if NrOf C opies > 1then
15: SendPacket(pkt,a)
16: StorePacket(pkt,b)
17: SetNrOfCopies(pkt,a,N rOfCopies/2)
18: SetNrOfCopies(pkt,b,N rOfCopies/2)
19: end if
20: end if
21: ContactDuration=ContactDuration-size(pkt)
22: end if
23: end if
24: end if
•Direct update:
P(a,b)=Pold
(a,b)+ (1 −Pold
(a,b))Pinit (1)
where Pold
(a,b)is the value of P(a,b)before updating,
Pinit ∈[0,1] is an initialization constant. This update
is done when the two nodes aand bcome into direct
contact with each other.
•Transitive update:
P(a,b)=Pold
(a,b)+ (1 −Pold
(a,b))P(a,c)P(c,b)β(2)
where β∈[0,1] is the transitivity constant which reflects
the impact of transitivity on the delivery predictability.
This equation updates the delivery predictability of node
atowards node athrough the transitive contact between
aand c.
•Aging:
P(a,b)=Pold
(a,b)γk(3)
where γ∈[0,1] is the aging constant, and kis the number
of time units that have elapsed since the last time the
metric was aged. This equation decreases the delivery
predictability by the time passed without direct between
the two nodes aand b.
PROPHET provides a partial guiding towards the desti-
nation by tracing the contacts between nodes and assigning
weights to these contacts whether they were directly or through
intermediate nodes. Therefore, PROPHET is expected to out-
perform the blind protocols in the delivery ratio. On the other
hand, it is expected that the average packet delay may increase
due to waiting for for a good next node in the path. A pseudo-
code for PROPHET is provided in algorithm 3.
4
Algorithm 3 PROPHET Routing Protocol
1: Procedure Name: OnContact
2: Input: node a, node b, integer ContactDuration
3: DropExpiredPackets(a,b) /* Drop packets with their life-
time expired in both nodes */
4: ExchangeSummaryVector(a,b)
5: UpdateDeliveryPredictability()
6: if ContactDuration > 0then
7: pkt=GetPacket(a)
8: if pkt then
9: if NotReceivedBefore(pkt,b)then
10: if IsDestination(pkt,b)then
11: SendPacket(pkt,a)
12: ConsumePacket(pkt,b)
13: else
14: DP n1=DeliveryPredictability(pkt,a)
15: DP n2=DeliveryPredictability(pkt,b)
16: if DP n2> DP n1then
17: SendPacket(pkt,a)
18: StorePacket(pkt,b)
19: end if
20: end if
21: ContactDuration=ContactDuration-size(pkt)
22: end if
23: end if
24: end if
D. MAXPPROP Routing Protocol
The MAXPROP protocol proposed in [10] estimates a node
metric, P(a, b), similar to PROPHET. When two nodes aand
bmeet, they strengthen the link between each other by adding
a constant αwhich is set to equal 1 in the protocol. Then the
two nodes divide their delivery predictability towards all the
nodes including each other by 1 + αso that the sum of all
delivery predictability remains 1.
P(a,b)=Pold
(a,b)+α(4a)
P(a,c)=Pold
(a,c)/(1 + α)(4b)
where α∈[0,1] is the updating constant which is set to 1 in
their work, and cis every other node including b.
The node metric is used only when the hop count of the
packet is greater than a certain threshold. The main contribu-
tion of MAXPROP is in its buffer management. Packets are
sorted according to their hop count, if the hop count is below
a certain threshold. Otherwise, packets are sorted with their
delivery predictability. In this way, MAXPROP favors packets
with less hop count to spread in the network.
III. PERFORMANCE MET RI CS AND SIMULATION SETU P
To compare the performance of the heuristic protocols with
each other and with the performance results of the optimal
routing, we built a DTN simulator in MATLAB. The simulator
takes as inputs the starting times and durations of node
contacts. The optimal routing problem is solved using an open
source mixed integer linear programming package LPSOLVE
Algorithm 4 MAXPROP Routing Protocol
1: Procedure Name: OnContact
2: Input: node a, node b, integer ContactDuration
3: DropExpiredPackets(a,b) /* Drop packets with their life-
time expired in both nodes */
4: ExchangeSummaryVector(a,b)
5: UpdateDeliveryPredictability()
6: SortPackets() /* Using MAXPROP sorting criteria */
7: if ContactDuration > 0then
8: pkt=GetPacket(a)
9: /* pkt is the packet with the minimum hop count, or
higher delivery predictability */
10: if pkt then
11: if NotReceivedBefore(pkt,b)then
12: if IsDestination(pkt,b)then
13: SendPacket(pkt,a)
14: ConsumePacket(pkt,b)
15: else
16: SendPacket(pkt,a)
17: StorePacket(pkt,b)
18: end if
19: ContactDuration=ContactDuration-size(pkt)
20: end if
21: end if
22: end if
[15]. We conducted the experiments using real life movement
scenarios of pedestrians and vehicles in a city. These inputs
are generated and recorded using the ONE simulator [16].
A. Optimal Routing
Optimal routing assumes the availability of full knowledge
about the network. There is only a single copy of each mes-
sage. For each message, optimal routing finds the optimal route
that maximizes a certain utility or minimizes a defined cost.
The optimal routing formulation was proposed and explained
in [11]. We solve the optimization problem with two objective
functions:
•MINH: Minimize the total number of hops.
•MIND: Minimize the total end-to-end delay.
B. Performance Metrics
We consider three metrics to measure the performance of
the different protocols, which are:
•Delivery ratio, DR:
DR =
∑
n∈N
(Pdv)n
∑
n∈N
(Pg)n
(5)
where (Pdv)nis the number of packets delivered to
their destination node n, and (Pg)nis the number of
packets generated at their source node n. The delivery
ratio is, simply, the ratio of the packets delivered to those
generated in the network during the simulation time.
5
TABLE I
NET WOR K PARAMETERS
Parameter Pedestrians Vehicles
#Hosts 5,20,30,45,60 5,10,15,25,30
Speed 0.5-1.5 m/s 2,7-13.9 m/s
Movement ShortestPath MapBased Movement
Buffer capacity 2-10 MBytes
Packet TTL 2-10 Hours
Packet Inter-arrival time 10,30,60,300,600 seconds
Transmission speed 5 MBps
Simulation time 12 Hours
•Packet delivery cost, DV C:
DV C =
∑
n∈N
(Pr)n−∑
n∈N
(Pdv)n
∑
n∈N
(Pdv)n
(6)
where (Pr)nis the number of packets received by node n.
DV C represents the cost paid using the routing protocol,
in terms of redundant packets, to deliver one packet.
•Average packet delay, Del:
Del =
∑
n∈N
∑
p∈DVn
dp
∑
n∈N
(Pdv)n
(7)
where dpis the delay encountered by packet pdelivered
to its destination node n, and DVnis the set of packets
delivered to their destination n. The metric is simply the
ratio of the sum of all delivered packets delays to the
number of delivered packets.
C. Network and Protocol Setup
The network consists of vehicles and pedestrians moving
around a city. We used the ONE simulator [16] to generate
the movement scenario. We set the mobility model as map-
based movement, where vehicles and pedestrians are restricted
to move in predefined paths and routes derived from real map
data. We used the map data of the Helsinki downtown area
(roads and pedestrian walkways) provided with the simulator.
We conducted four sets of experiments to study the impact
of the following parameters:
•Buffer Capacity,
•Packet lifetime or time-to-live (TTL),
•Node density by changing the number of nodes in the
network, and
•Traffic load by changing the packet generation rate
on the performance of four DTN routing protocols: EPI-
DEMIC, Spray-and-Wait (SnW), PROPHET, and MAXPROP.
We used the optimal parameter values specified in [9] for the
PROPHET protocols. We used five packet copies for SnW.
Table I shows the network parameters used in the experiments.
IV. SIMULATION RESULTS
Table II shows the summary of the results obtained from all
the experiments. We presented the minimum and maximum
values for the three performance metrics in the five experi-
ments. Delivery cost is measured in number of packets, and
average delay is measured in minutes. Protocols are ordered
from the highest to lowest in delivery ratio, and from lowest to
highest in other metrics. The detailed experiments are shown
in the following subsections.
A. Impact of Varying the Buffer Capacity (B)
From Figure 1, it can be observed that increasing the buffer
capacity increases the delivery ratio of all the protocols, as
long as the amount of bytes of the propagating packets are
more than the buffer space. The delivery ratio settles when the
buffer space is larger than that of the propagating data, i.e. no
packets are dropped because of buffer overflow. The delivery
ratio of both optimal routing protocols settles at a small value
of the buffer capacity because it propagates only a single
copy of each packet. At very small buffer capacity, MIND
outperforms MINH because it favors routes with lower delay,
which helps in flushing buffers before packets got dropped.
It is also noticed that among the distributed heuristic (non-
optimal) protocols, MAXPROP protocol provides the highest
delivery ratio, and Epidemic Routing provides the lowest one.
Fig. 1. Impact of changing buffer capacities on Delivery Ratio.
The high delivery ratio for MAXPROP is achieved with
a large cost (network overhead) of delivering a packet, as
shown in Table II. The large network overhead also represents
high energy consumption during transmissions and receptions.
Both optimal objectives guarantee the least cost because they
only propagate a single packet copy. The low cost of SnW,
despite its blind routing as EPIDEMIC, is because of its partial
flooding. The highest cost is for EPIDEMIC because of its full
flooding behavior.
B. Impact of Varying Packet Lifetime (TTL)
Increasing the lifetime (TTL) of data packets, increases the
delivery ratio up to a maximum value, as shown in Figure
2. On the other hand, it overloads the buffer space available
which may lead to an increase in dropping the stored packets.
Overloading effect is significant in the case of EPIDEMIC
where delivery ratio is found to be decreasing at values
T T L > 4hours as a result of the increased dropping because
of buffer overflow.
6
TABLE II
SUM MARY O F PERFORMANCE COMPA RIS ON O F DTN ROUTING PROTO COL S:
EPIDEMIC ROUT ING [7], PROPHET [9], SNW [8], MAXPROP [10], MINH AND MIND
Impact of Buffer TTL N TL Comments
MIND 0.83-0.90 MIND 0.79-0.90 - Ratio of the number of
MINH 0.74-0.90 MINH 0.79-0.90 packets delivered to the
Delivery MAXPROP 0.57-0.90 MAXPROP 0.80-0.90 MAXPROP 0.90-0.97 MAXPROP 0.43-0.98 number of packets generated
Ratio PROPHET 0.45-0.81 SnW 0.78-0.81 SnW 0.81-0.94 SnW 0.33-0.95 - Measured as a fraction
SnW 0.42-0.81 PROPHET 0.69-0.82 PROPHET 0.82-0.88 PROPHET 0.26-0.95 - Ordered from high to low
EPIDEMIC 0.36-0.72 EPIDEMIC 0.74-0.80 EPIDEMIC 0.74-0.77 EPIDEMIC 0.74-0.77 - The higher the better
Numbers shown are the minimum and maximum values
MINH 0.21-0.23 MINH 0.22-0.44 - Ratio of the amount of
MIND 0.94-0.97 MIND 0.86-0.94 redundant data to the
Delivery PROPHET 3.1-3.5 PROPHET 2.8-3.5 SnW 4.2-5.9 SnW 4.6-5.8 amount of delivered data
Cost SnW 3.7-4.2 SnW 4.2-4.4 PROPHET 3.7-7.0 MAXPROP 28-36 - Measured in packets
MAXPROP 4.5-5.1 MAXPROP 5.1-5.3 MAXPROP 5.1-6.7 PROPHET 20-29 - Ordered from low to high
EPIDEMIC 5.6-6.9 EPIDEMIC 6.5-6.8 EPIDEMIC 6.8-8.6 EPIDEMIC 36-46 - The lower the better
Numbers shown are the minimum and maximum values
MIND 58-59 MIND 46-59 - Average delay of delivered
MAXPROP 58-63 MAXPROP 45-58 MAXPROP 18-58 MAXPROP 24-60 packets
Average SnW 62-72 SnW 46-62 EPIDEMIC 19-62 EPIDEMIC 24-72 - Measured in minutes
Delay EPIDEMIC 63-73 EPIDEMIC 45-62 SnW 39-62 SnW 40-81 - Ordered from low to high
PROPHET 72-75 PROPHET 46-72 PROPHET 24-72 PROPHET 39-82 - The lower the better
MINH 108-118 MINH 54-116
Numbers shown are the minimum and maximum values.
Fig. 2. Impact of changing packet TTL on Delivery Ratio.
C. Impact of Varying the Number of nodes (N)
In these experiments, we did not solve the optimal routing
because of its long processing times. Therefore, we compare
only the heuristic protocols. Increasing the number of nodes,
while fixing the packet generation rate, improves the con-
nectivity of the network nodes and allows for more packets
to be delivered, as shown in Figure 3, but with increased
number of hops. Increasing the number of nodes allows also
for increased number of packet copies to be generated, as
shown in Table II, unless the protocol limits the number of
copies as in SnW. Increasing the node density provides faster
paths to destinations which decreases the average packet delay.
D. Impact of Varying the traffic load (TL)
In this experiment, we ran the simulation five times with
different packet generation rates. The inter-generation times
are drawn uniformly from the intervals: 8-12, 25-35, 55-
65, 250-350, and 550-650 seconds. Therefore, the average
generation rates are found to be 379, 121, 66, 12, and 6
Fig. 3. Impact of changing number of nodes on Delivery Ratio.
packets/hour respectively. Increasing the traffic load, in a
network with fixed number of nodes, causes the overloading of
buffers and increases the dropping rate. Therefore, the delivery
ratio drops significantly, see Figure 4. Packet propagation
decreases in the network because many packets are dropped,
which maintains an almost non-varying delivery cost, with an
increasing end-to-end delay, as shown in Table II.
V. DESIGN GU ID EL IN ES O F DTN ROU TI NG PROTOCOLS
From the analysis of the routing protocols and the simu-
lation results (see Table II), we come out with the following
points:
•Guided-Routing protocols, which have a node selection
mechanism, such as MAXPROP and PROPHET, outper-
form Blind-Routing protocols in the delivery ratio. How-
ever, they increase the end-to-end delay, as in PROPHET,
unless they have an efficient packet selection mechanism
as in MAXPROP.
7
Fig. 4. Impact of changing the traffic load on Delivery Ratio.
•Full-Flooding protocols, such as EPIDEMIC, consume
huge resources, such as node buffer, energy and contact
durations. Therefore, they have the highest delivery cost.
•Partial-Flooding protocols, such as SnW, achieve better
delivery cost and have an improved delivery ratio and
average delay compared to other protocols.
Therefore, we outline the key design issues of an efficient
DTN routing protocol as follows:
•Exploit the available, easily and quickly collected, in-
formation in the network to guide the packets to their
destinations. The information could be related to topology
[9], [12], energy content [17], data content [18], or social
relations [19].
•Limit the number of copies of each packet in the network
to decrease the network overhead and preserve node
energies.
•Implement an efficient packet selection mechanism that
reduces the end-to-end delay.
•Implement an efficient buffer management mechanism
that can free space in the buffers by dropping packets
that are less probable to be delivered.
•Integrate the node selection, the packet selection, and
the buffer management mechanisms to obtain an efficient
DTN routing protocol.
VI. CONCLUSION
Delay Tolerant Networks (DTN), lack end-to-end connec-
tions between source and destination nodes. This environment
requires the intermediate nodes to store data packets for long
periods of time which violates one of the basic assumptions of
traditional routing protocols and triggers the development of
new ones. DTN routing protocols vary according to the amount
of information they acquire to take the routing decision. Blind-
Routing protocols do not collect any information about the
network and, therefore, they do not have a node selection
mechanism. They just spread the packets so that one of
the copies might reach the destination. The performance is
improved if the packet spreading is limited. Guided-Routing
protocols seek the possible paths to destinations, by selecting
the relay nodes, which improves the delivery ratio. A packet
selection mechanism helps in reducing end-to-end delays.
A buffer management mechanism helps in providing buffer
space for newly generated and arriving packets. In this paper,
we presented and compared four well-known DTN routing
protocols together with the optimal routing results. EPIDEMIC
is an example of a Blind-Routing Full-Flooding protocol. SnW
is an example of a Blind-Routing Partial-Flooding protocol.
PROPHET is an example of a Guided-Routing protocol with
a first in first out (FIFO) packet selection mechanism. MAX-
PROP is an example of a Guided-Routing protocol which
favors packets with minimum number of hops. We conducted
simulations using real life scenarios of vehicles and pedes-
trians roaming in a city. Results show the outperformance
of MAXPROP in delivery ratio and average delay, and the
outperformance of SnW and PROPHET in delivery cost. An
efficient routing protocol should integrate the node selection,
the packet selection, and the buffer management mechanisms
to obtain the best performance.
REFERENCES
[1] “Delay tolerant networking research group.” [Online]. Available:
http://www.dtnrg.org
[2] A. Pentland, R. Fletcher, and A. Hasson, “Daknet: rethinking connec-
tivity in developing nations,” Computer, vol. 37, no. 1, pp. 78 – 83, jan.
2004.
[3] N. Benamar, K. D. Singh, M. Benamar, D. El Ouadghiri, and J.-M.
Bonnin, “Routing protocols in vehicular delay tolerant networks: A
comprehensive survey,” Computer Communications, vol. 48, pp. 141–
158, 2014.
[4] J. Ott and D. Kutscher, “Drive-thru internet: Ieee 802.11b for automobile
users,” in INFOCOM 2004. Twenty-third Annual Joint Conference of the
IEEE Computer and Communications Societies, march 2004.
[5] P. Juang, H. Oki, Y. Wang, M. Martonosi, L. S. Peh, and D. Rubenstein,
“Energy-efficient computing for wildlife tracking: design tradeoffs and
early experiences with zebranet,” SIGARCH Comput. Archit. News,
vol. 30, pp. 96–107, October 2002.
[6] “Sensor networking with delay tolerance (sendt).” [Online]. Available:
http://down.dsg.cs.tcd.ie/sendt
[7] A. Vahdat and D. Becker, “Epidemic routing for partially connected ad
hoc networks,” in Technical Report CS-200006, Duke University, april
2000.
[8] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Efficient routing
in intermittently connected mobile networks: the multiple-copy case,”
IEEE/ACM Trans. Netw., vol. 16, no. 1, pp. 77–90, 2008.
[9] A. Lindgren, A. Doria, and O. Schelen, “Probabilistic routing in in-
termittently connected networks,” SIGMOBILE Mobile Computing and
Communication Review, vol. 7, no. 3, pp. 19–20, July 2003.
[10] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine, “Maxprop:
Routing for vehicle-based disruption-tolerant networks,” in INFOCOM
2006. 25th IEEE International Conference on Computer Communica-
tions. Proceedings, april 2006, pp. 1 –11.
[11] T. Abdelkader, K. Naik, and A. Nayak, “Choosing the objective of
optimal routing protocols in delay tolerant networks,” in Computer
Engineering Conference (ICENCO), 2010 International, pp. 16-21, Dec.
2010.
[12] T. Abdelkader, K. Naik, A. Nayak, and N. Goel, “A socially-based rout-
ing protocol for delay tolerant networks,” In Global Telecommunications
Conference (GLOBECOM 2010), 2010 IEEE, pp. 1-5.
[13] Z. Zhang, “Routing in intermittently connected mobile ad hoc networks
and delay tolerant networks: overview and challenges,” Communications
Surveys & Tutorials, IEEE, vol. 8, no. 1, pp. 24–37, March 2007.
[14] 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, Dec. 2013.
[15] “Lpsolve: Open source (mixed-integer) linear programming system.”
[Online]. Available: http://sourceforge.net/projects/lpsolve/
[16] A. Ker¨
anen, J. Ott, and T. K¨
arkk¨
ainen, “The ONE Simulator for
DTN Protocol Evaluation,” in SIMUTools ’09: Proceedings of the 2nd
International Conference on Simulation Tools and Techniques. New
York, NY, USA: ICST, 2009.
8
[17] H. Jun, M. H. Ammar, M. D. Corner, and E. W. Zegura, “Hierarchical
power management in disruption tolerant networks with traffic-aware
optimization,” in Proceedings of the 2006 SIGCOMM Workshop on
Challenged Networks, ser. CHANTS ’06. New York, NY, USA: ACM,
2006, pp. 245–252.
[18] P. Costa, C. Mascolo, M. Musolesi, and G. Picco, “Socially-aware
routing for publish-subscribe in delay-tolerant mobile ad hoc networks,”
Selected Areas in Communications, IEEE Journal on, vol. 26, no. 5, pp.
748–760, June 2008.
[19] K. Wei, X. Liang, and K. Xu, “A survey of social-aware routing protocols
in delay tolerant networks: Applications, taxonomy and design-related
issues,” Communications Surveys Tutorials, IEEE, vol. 16, no. 1, pp.
556–578, First 2014.
Tamer Abdelkader received the B.Sc. degree in
electrical and computer engineering and the M.Sc.
degree in 2003 in computers and information sci-
ences, from Ain Shams University, Cairo, Egypt. He
received M.Sc. and Ph.D. degrees in 2012 in elec-
trical and computer engineering from the University
of Waterloo, Ontario, Canada. He worked in the uni-
versity of Waterloo as a postdoctoral researcher, and
a consultant in the Information and Communication
Technology Project (ICTP) in Egypt. Currently, he
is an assistant professor in the Information Systems
Department at Ain Shams University. He is the author of several publications
in IEEE journals and conferences. His current research interests include re-
source allocation in wireless networks, mobile computing, vehicular networks,
energy efficient protocols, selfishness in mobile networks and data mining in
mobile networks.
Kshirasagar (Sagar) Naik received the B.Sc. En-
gineering degree from Sambalpur University, India
and M. Tech. degree from Indian Institute of Tech-
nology, Kharagpur, respectively. He worked as a
software developer in Wipro Information Technol-
ogy Limited, Bangalore for three years. Next, he
received the M. Math degree in computer science
from University of Waterloo and Ph.D. degree in
electrical and computer engineering from Concordia
University, Montreal, respectively. He worked as a
faculty member at the University of Aizu in Japan
and Carleton University in Ottawa. Currently, he is a full professor in
the Department of Electrical and Computer Engineering at the University
of Waterloo. He was a co-guest editor of three special issues of IEEE
Journal on Selected Areas in Communications. He was an associate editor of
Journal of Peer-to-Peer Networking and Applications, International Journal
of Parallel, Emergent and Distributed System, and IEEE Transactions on
Parallel and Distributed Systems. Now he is serving as a Regional Editor
(America) of Journal of Circuits, Systems, and Computers and an Associate
Editor of International Journal of Distributed Sensor Networks. His research
interests include dependable wireless communication, resource allocation in
wireless networks, mobile computing, vehicular networks, energy efficiency
of smartphones and tablet computers, energy performance testing of mobile
apps, communication protocols for smart power grids, and energy performance
testing of software systems running on servers. He was a recipient of
the Outstanding Performance Award in the Faculty of Engineering at the
University of Waterloo in 2013. His book entitled Software Testing and
Quality Assurance: Theory and Practice (Wiley, 2008) has been adopted as a
text in many universities around the world, and he has co-authored a second
book entitled Software Evolution and Maintenance (Wiley, 2014).
Amiya Nayak has 17 years of industrial experience
in software engineering, avionics and navigation sys-
tems, and system level performance analysis. Now
is a full professor in the School of Information
Technology and Engineering at the University of
Ottawa. He is on the Editorial Board of IEEE
Transactions on Parallel and Distributed Systems.
His research interests are in the areas of mobile ad-
hoc and sensor networks, fault tolerance, dependable
communication, intelligent transportation systems,
and distributed systems, with over 150 publications
in refereed journals and conference proceedings. He has co-edited Handbook
of Applied Algorithms: Solving Scientific, Engineering, and Practical Prob-
lems (Wiley, 2008) and co-authored Wireless Sensor and Actuator Networks:
Algorithms and Protocols for Scalable Coordination and Data Communication
(Wiley, 2010).
Nishith Goel is the CEO of Cistel Technology,
an Information Technology company he founded in
1995 which has operations in Canada and the U.S.
A veteran technology executive and entrepreneur,
Dr. Goel is also co-founder of CHiL Semiconductor,
IPine Networks and Sparq Systems. After his early
education in Jodhpur, India, Dr. Goel earned his
MASc in Electrical Engineering and PhD in Systems
Design Engineering from the University of Waterloo.
He began his professional career at Bell Northern
Research in Ottawa in 1984 before moving on to
Northern Telecom in 1988. His research interests are in information technol-
ogy, wireless networks and network security. He serves on various corporate
boards of directors and is currently Chair of Queen University Center for
Energy and Power Electronics Research.
Vineet Srivastava is currently a Chief Operating Officer at Cistel Technology.
Cistel provides a wide range of IT services and employs over 170 employees
and associates in Canada and the US. Vineet is a Mechanical Engineer by
training, but has spent last 23 years in the Information Technology space.
At Nortel, Vineet worked in various roles as an individual contributor and
as senior management and won several recognitions, including top-talent.
Vineet joined Cistel as Vice President in 2001 with an entrepreneurial hat.
During the last 8 years, revenues have grown four-folds to over $12M in 2008
and upwards of 130 new jobs created. The key guiding principles Vineet
has adopted in his approach include wealth creation via steady sustainable
growth, long-term stability of the business, and profitability. After receiving
his Bachelor’s degree in Mechanical Engineering from Birla Institute of
Technology (India) in 1985, Vineet went to University of Alaska at Fairbanks
where he received his Master’s degree in Mechanical Engineering in 1987
and a MBA in 1990.