Simulation of Mobility and Routing in Ad Hoc Networks using Ant Colony Algorithms

elfinoverwroughtNetworking and Communications

Jul 18, 2012 (5 years and 3 months ago)

369 views

Simulation of Mobility and Routing in Ad Hoc Networks using Ant Colony
Algorithms
Tarek H. Ahmed
Institute for Computer Science and Business Information Systems
University of Duisburg-Essen, Germany
E-mail: tarek@informatik.uni-essen.de
Abstract
Mobile Ad-hoc Networks (MANET’s) have
recently attracted a lot of attention in the research
community as well as the industry. This technology
has become increasingly important in
communication and networking. Routing is one of
the most important and difficult aspects in ad hoc
network since ad hoc network topology frequently
changes. Conventional routing algorithms are
difficult to be applied to a dynamic network
topology, therefore modeling and design an efficient
routing protocol in such dynamic networks is an
important issue. One of the meta-heuristic
algorithms which is inspired by the behavior of real
ants is called Ant Colony Optimization (ACO)
algorithm, it can definitely be used as a tool to tackle
the mercurial scenarios present in this dynamic
environment. In this paper, I have designed a model
which combines ant colony behavior and queuing
network analysis to evaluate End-to-End packet
delay in MANET.
1. Introduction
Mobile ad hoc network (MANET), or simply ad
hoc network,is one of the most innovative and
challenging areas of wireless networking, one which
promises to become increasingly present in our lives.
Ad hoc network consists of nodes that are freely and
dynamically self-organized (i.e. nodes are
autonomous) into arbitrary and temporary network
topology without any infrastructure support [5].
Nodes are computing and communication devices
which can be laptop computers, PDAs, mobile
phones or even sensors. Ad hoc networks offer a
large degree of freedom at a lower cost than other
networking solutions. The ease and speed of
deployment of these networks make them ideal
for disaster recovery (such as a hurricane,
earthquake or flooding), business associates
sharing information during meeting, conferencing,
and military communications as in a battlefield.
One of the main problems in mobile ad hoc
networks is to find a best route between the
communication end-points, which may be difficult
due to node mobility. Several routing algorithms
have been proposed in the literature, some of these
are DSDV,AODV,DSR,TORA and several others
[6]. The goal of every routing algorithm is to direct
traffic from sources to destinations, maximizing
network performance while minimizing cost.
In this paper, I will present a new approach for an
ad hoc routing algorithm, which is based on Ant
Colony Optimization (ACO) algorithm and its
combination with queuing network analysis. The
basic idea of the ACO meta-heuristic is taken from
the food searching behavior of real ants. While
walking, ants deposit pheromone, which marks the
route taken as they move from a food source to
their nest, and foragers follow such pheromone
trails. The concentration of pheromone on a certain
path is an indication of its usage. These pheromone
trails are used as a simple indirect form of
communication. The process of emerging global
information from local actions through small, in-
dependent agents not communicating with each
other is called Stigmergy [2]. This behavior of the
ants can be used to find the shortest path in networks.
Especially, the dynamic component of this method
allows a high adaptation to changes in mobile ad hoc
network topology, since in these networks the
existence of links are not guaranteed and link
changes occur very often.
The simple ant colony optimization meta-heuristic
illustrates why this kind of algorithms could perform
well in mobile ad hoc networks by different reasons.
The main reason is that ACO meta-heuristic is based
on agent systems and works with individual ants.
This allows a high adaptation to the current topology
of the network. One other reason is the way of taking
decision about selecting the next node that is based
on the pheromone concentration on the current node
Proceedings of the International Conference on Information Technology: Coding and Computing (ITCC’05)
0-7695-2315-3/05 $ 20.00 IEEE
which is provided for each possible link. Thus, the
approach supports multipath routing
[7]
. In this
paper, I present the ACO principle in a unifying
framework which also includes mobility modeling
and queuing network analysis to evaluate End-to-
End packet delay in MANET.
2. Previous work
A great deal of literature has been published in the
field of mobile ad hoc networks. Early proposals
optimize for traditional metrics such as path length or
energy use. Recently, more efforts have been spent to
use specific network parameters when specifying
routing metrics. There exists relatively little work
with regards to biologically inspired algorithms for
routing in communications networks. However, there
are a number of notable examples which show that
these concepts can provide a significant performance
gain over traditional approaches. The following
paragraphs describe a number of the most prominent
algorithms in this field.
2.1. Ant-Based Control
Ant-Based Control (ABC) is a routing algorithm
for circuit-switched networks in which routes calls
are based on the local interaction of mobile agents
[9]. Mobile agents (ant packets) traverse the network,
updating routing tables at each node depending on
the current state of the routing table as well as the age
of the packet. Routing tables consist of next hop
probabilities for each destination. Ants traveling in
one direction influence the placement of calls in the
opposite direction.
2.2. AntNet
AntNet is an adaptive routing algorithm inspired
by ant colonies to solve routing problems in wired
networks [3]. An AntNet node maintains
probabilistic entries in the routing table, indicating
the goodness of each output link for each destination.
Each node periodically sends a forward ant packet
to explore paths to a random destination. Forward
ants explore the network to find a feasible and low-
cost path, recording every node it visits. Once it
arrives at the destination, it is converted into a
backward ant. The backward ant returns to the source
node following the path in reverse. Each
intermediate node updates its routing tables with
the information from the backward ant. Ants
interact and communicate indirectly by updating the
routing tables, thus collaboratively solve the global
network routing optimization problem.
2.3. Mobile Ants Based Routing
Mobile Ants Based Routing (MABR) is
introduced as the first routing algorithm for
MANET’s inspired by social insects [8]. The
approach presented in AntNet is extended to ad hoc
networks by abstracting the network into logical
links and nodes based on relative node location.
Location data is assumed from positioning
devices. An optimized greedy routing algorithm is
used to forward messages between logical nodes.
2.4. Ant Colony Based Routing Algorithm
This algorithm (ARA) presents a detailed routing
scheme for MANET’s, including route discovery
and maintenance mechanisms [7]. Route discovery
is achieved by flooding forward ants to the
destination while establishing reverse links to
the source. A similar mechanism is employed in
other algorithms such as AODV. Routes are
maintained primarily by data packets as they flow
through the network. In the case of a route failure,
an attempt is made to send the packet over an
alternate link. Otherwise, it is returned to the
previous hop for similar processing. If the packet is
eventually returned to the source, a new route
discovery sequence is launched.
3. Simulation algorithm
In this section, I will describe in details the
proposed framework for routing algorithm based on
ACO algorithms and Kleinrock’s delay analysis to
find the best route with minimum End-to-End packet
delay in a MANET. Figure 1 illustrates a complete
scenario of the simulation process using both ant and
delay analysis algorithms.
3.1. The initialization step
3.1.1. Mobility models. The simulation environment
consists of a set of mobile nodes (MN’s) with bi-
directional wireless links that are moving on the
simulation area (rectangular area) according to one of
two different mobility models, Random Waypoint
Mobility (RWM) model and Boundless Simulation
Area Mobility (BSAM) model [10].
RWM model states that in the beginning of the
simulation, each MN picks a random destination in
the area, traverses to that destination in a straight line
at a uniform speed, then each MN is staying in its
location for a certain period of time called pause
time. When this time does expire, each MN chooses a
new random destination, then they travel towards the
Proceedings of the International Conference on Information Technology: Coding and Computing (ITCC’05)
0-7695-2315-3/05 $ 20.00 IEEE
newly chosen destination by the same way and so on
as shown in Figure 2. RWM model attempts to make
the movement of the nodes more realistic by allowing
the nodes to remain stationary for a period of time
(the pause time) before moving to a next location.
Figure 1. Flowchart for complete scenario
In the BSAM a relationship between the previous
and current movement direction ș and velocity v
exists in order to limit the change in direction and
speed per time unit to generate more realistic
movement patterns. Both the velocity vector
V = (v,ș) and the MN’s position (x,y) are updated at
every ¨t time steps according to formulas in [10] to
generates a new topology. An other speciality of this
mobility model has given it its name, the rectangular
simulation area is folded to form a torus with the
effect that nodes moving out of the simulation area
enter it again at the opposite side thus creating a
boundless simulation area (e.g. a node leaving the
simulation area at the top enters at the bottom again)
as shown in Figure 3. Since the changes in direction
and speed are limited, the resulting moves lack of
abrupt changes in direction and speed, which makes
the movement patterns more realistic. This realism
only persists, if the components working with the
resulting movements are aware of the speciality of
the boundless simulation area.
Figure 2. RWM model
Figure 3. Rectangular simulation area
mapped to a torus in the BSAM model
3.1.2. Timers. We define two different timers at the
beginning of the simulation, T
Simul
is a timer for the
whole simulation time and T
ant
= pause time is the
time between changing from one network topology to
another topology.
3.1.3. Initial routing tables. Every node has a table
of next-hop probabilities to each destination. Each
row in this table corresponds to a destination and
each column corresponds to a neighbor within its
current transmission range. The entries in the table
are the probabilities of taking a next-hop at a certain
node to eventually reach a certain destination.
3.2. Kleinrock’s delay analysis
We sketch how End-to-End delay through the
network can be calculated using Kleinrock’s
independence assumption. The expected response
time (delay) to send packets from a source node S to
a destination node D is the sum over the response
times at all links and nodes visited along the way [1]:
250 m
Closed Coverage
Area
X
max
, Y
max
X
max
, 0
0, Y
max
0, 0
Each forward ant
move after using
R
oulette Wheel
selection on the
new probabilities
ij
p

Start
Initial network topology,
Initial routing table with probabilities P
ij
,
Simulation-Timer T
simul
, Ant-Timer T
ant
Initial parameters Ȝ ,
µ
i
,B
ij
E[V] calculated by
QN-algorithms
Ant Algorithm
New Routing tables
probabilities
ij
p

E[V] calculated by
QN-algorithms
N
ew E[V]
less than
Old E[V]
?
End of
T
ant
?
End of
T
ant
?
Each forward ant
move after using
R
oulette Wheel
selection on the
old probabilities
P
ij
End of T
simul
?
Random Waypoint
Mobility Model
N
ew Network
Topology
End
N
o
N
o
N
o
Yes
Yes
Yes
Yes
N
o
Proceedings of the International Conference on Information Technology: Coding and Computing (ITCC’05)
0-7695-2315-3/05 $ 20.00 IEEE

][][)],([
 
+=
nij
REREDSRE
Where
ijij
ij
C
RE
λµ −
=
1
][
, is the expected delay at
all links,
µ
C
ij
is the number of packets that can be
transmitted over a link ij (packets/sec.), and
λ
ij
is the
arrival rate over a link ij.
][
1
][
][
n
n
nn
n
SE
SE
RE +

=
ρ
ρ
, is the
expected delay at n nodes with E[S
n
]=1/
µ
n
and
ρ
n
=
λ
n
/
µ
n
.
3.3. Ant routing algorithm
The work reported here follows the AntNet
algorithm [4], and is informally summarized as
follows:

Each node in the network retains a record of packet
destinations as seen on data packets passing
through that node. This is used to periodically, but
asynchronously, launch ‘forward’ ants with
destinations stochastically sampled from the
collected set of destinations.

Once launched, a forward ant uses the routing table
information to make probabilistic decisions
regarding the next hop to take at each node. While
moving, each forward ant collects time stamp and
node information in a stack, which is later used to
update the routing tables along the path followed.
The trip time to reach the desired neighbor is
computed using this simple formula:
d
ij
+ (q
ij
+ S
a
)/ B
ij
(1)
Where d
ij
is the link’s propagation delay
(distance/signal propagation speed) between two
mobile nodes i and j. Note that this value can be
neglected because the distance value is very small
in comparison to the value of signal propagation
speed.
The number of data packets waiting in the queue
between nodes i and j is q
ij
and is calculated by
using the M/M/1 equation
ij
ij
ij
q
ρ
ρ

=
1
2
,
where
j
ijj
ij
P
µ
λ
ρ
×
=
is the utilization of the link
between two nodes i and j, and
λ
j
and
µ
j
are the
arrival rate and service rate at node j respectively.
P
ij
is the probability of routing from node i to j.
S
a
is the size of the ant packet, and B
ij
is the
bandwidth of the link between two nodes i and j.
Bandwidth is here defined as the amount of data
that can be transmitted in a fixed amount of time,
expressed in bits per second (bps).
Equation (1) represents the time delay for links.
We suppose that the node delay is also determined
by M/M/1 formulas for all nodes;

If a forward ant re-encounters a node previously
visited before reaching the destination, it is killed
(in other words, identification of a loop in the
path);

On successfully reaching the destination node, the
total trip time is calculated, and the forward ant is
converted into a backward ant;

The backward ant returns to the source using
exactly the same route recorded by the forward ant.
Instead of using the data packet queues, the
backward ant uses priority queues;

At each node visited by the backward ant, the
corresponding routing table entries are updated to
reflect the relative performance of the path. The
backward ant retraces the path of the forward
ant by popping the stack, making
modifications in the routing tables at each
intermediate node according to the following
learning rules:
IF
(node was in the path of the ant)
THEN
p(i) = p(i) + r [1-p(i)]
ELSE
p(i) = p(i) - r p(i)
Where r

(0,1] is the reinforcement factor central
to express path quality. The reinforcement factor
should be a factor of trip time. This factor is given
by the following relationship, r=t
1
/t
2
where t
1
is the
minimum trip time of all the forward ants, and t
2
is
the trip time of the current forward ant from a node
to the destination node.

When the backward ant reaches the source, it dies.
My simulation model can be summarized as follows:
Given a starting topology, then its delay is calculated.
After applying the ant algorithm on the current
topology, the probabilities in routing tables are
altered a little bit and then, the delay is recalculated.
If the new delay is better, it definitely becomes the
starting point for the next iteration and each forward
ant will move by applying roulette wheel selection on
the new probabilities after updating. If the delay is
worse each forward ant will move by applying
roulette wheel selection on the old probabilities
before updating. This has the effect that, nearly every
new solution is adopted, while over time it becomes
more and more likely that only better solutions are
accepted.
This comparison between the delay of each
successive iterations repeats until the ant timer T
ant
finishes, then if the simulation timer T
Simul
hasn’t
finished yet, a new topology of the network will be
established according to one of the mobility models
RWM model or BSAM model and repeats again the
same procedure until the simulation timer finishes.
Proceedings of the International Conference on Information Technology: Coding and Computing (ITCC’05)
0-7695-2315-3/05 $ 20.00 IEEE
4. Performance evaluation
In this section we evaluate the End-to-End delay
E[V] for the packets through the MANET using the
framework described in the preceeding section.
4.1.Simulation model
Simulation scenario consists of a number of nodes
that are initially placed randomly and constantly
moving in a simulation rectangular area 1000
×
800
m
2
according to RWMand BSAMmodels.
During the simulation by RWM model, nodes are
free to move anywhere within this area. Each node
travels towards a random spot, then take a rest period
of time in second. After the rest period, the node
travels towards another randomly selected spot. On
the other hand, in case of BSAM model, all MN’s are
moving according to two parameters, velocity v as
well as direction ș of the MN’s.
The movements process in both mobility models
RWM and BSAM repeats throughout the simulation,
causing continuous changes in the topology of the
underlying network, followed by a simulation of the
ant behaviour yielding an improvement of the routing
tables, which is evaluated by Kleinrock´s delay
analysis. Finally, we get the minimum delay from
source to destination node. The simulation program
has been executed on standard 350 Mhz PC using
Visual Basic 6.0. It needs few seconds of CPU time
for a single simulation run.
The model parameters that have been used in the
following experiments are summarized in Table1.
Table 1. Simulation parameters
Parameter
Value
Number of nodes 10, 20, …, 150
Arrival rate 150 Kbps
Transmission range 250 m
Velocity / Direction 10 m/sec & 45 degree
Packet size 64 byte
Pause time (Ant time) 10, 20, …, 100 sec.
Simulation time 180 sec.
Link bandwidth 1 Mbps
Simulation area 1000 m × 800 m
Mobility model RWM & BSAM models
Routing protocol Ant algorithm + Delay
Analysis by Kleinrock
4.2. Results
The first experiment shows the relation between
increasing the number of nodes and the End-to-End
delay in case of using different numbers of nodes
(10, 20, … , 150) while ant time is constant at 30 sec.
Figure 4 shows that increasing the number of nodes
results in an increase in the delay, because each hop
can contribute a substantial amount of delay in
forwarding traffic. Furthmore, the more nodes, the
more congestion, and the longer it takes to discover
routes.
Ant time = 30 sec
0.00
0.10
0.20
0.30
0.40
0.50
0.60
10
30
50
70
90
110
130
150
Number of nodes
Mean delay
[sec]
Figure 4. Number of nodes vs. delay
The second experiment investigates the relation
between increasing the ant time and the End-to-End
delay in case of using different pause times (10, 20,
…, 100 seconds), and 60 nodes.
Number of nodes = 60
0.068
0.072
0.076
0.080
0.084
0.088
10 20 30 40 50 60 70 80 90 100
Ant time [sec]
Mean delay [sec]
Figure 5. Ant time vs. delay
Figure 5 shows that increasing the pause time
leads to a decrease in the delay, because the ant
algorithm performs more iterations which help to
approach the minimum delay.
The third experiment is done with RWM and
BSAM models for a scenario with 60 nodes
distributed randomly, and an ant time of 60 seconds.
The whole simulation time is 180 seconds and we
observe the packet delays over this time period.
Number of nodes = 60
0.076
0.078
0.080
0.082
0.084
0.086
0.088
5 10 15 20 25 30 35 40 45 50 55 60
Ant time [sec]
Mean delay [sec]
Figure 6. Mean delay through 1st
topology
Proceedings of the International Conference on Information Technology: Coding and Computing (ITCC’05)
0-7695-2315-3/05 $ 20.00 IEEE
Figure 6 shows the mean delay for the first topology.
Figures 7 and 8 show the mean delay for the
topology which has been established from the
preceding topology by applying the mobility model.
The following figures do display 95% confidence
intervals.
Number of nodes = 60
0.076
0.078
0.080
0.082
0.084
0.086
65 70 75 80 85 90 95 100 105 110 115 120
Ant time [sec]
Mean delay [sec]
Figure 7. Mean delay through 2nd
topology
Number of nodes = 60
0.068
0.070
0.072
0.074
0.076
0.078
0.080
0.082
125 130 135 140 145 150 155 160 165 170 175 180
Ant time [sec]
Mean delay [sec]
Figure 8. Mean delay through 3rd
topology
Figure 9 summarizes the tracing of the delay over
the whole simulation time of 180 seconds for three
different topologies. It is clear that, there is no
significant effect on the delay between RWM and
BSAM models, but the performance can vary
significantly with other different mobility models as
stated in [10]. After a change in the topology, we
observe a performance improvement due to the work
of the ant algorithm which converges step-by-step
towards the minimum possible delay.
0.065
0.07
0.075
0.08
0.085
0.09
10
30
50
70
90
110
130
150
170
Ant time [sec]
Mean delay [sec]
BSAM
RWM
Number of nodes = 60
Figure 9. Packet delay over 180 sec.
5.Conclusion and future work
A routing algorithm for mobile wireless ad-hoc
networks has been presented together with a hybrid
simulation framework which has been established to
perform model experiments that give insight into the
behavior of the proposed ant routing protocol.
The simulation experiments showed that the
considered ant algorithm is able to cope with this
type of dynamic networks, in particular its ability to
improve the system performance which has been
reflected in the model. It is also showed that, there is
no significant effect on the delay between RWM and
BSAM models.
Experiments with other scenarios (e.g. other
mobility models, other traffic models using multiple
source nodes and multiple destination nodes) are
necessary to show its general applicability. In
particular the comparison with other routing
algorithms will be in the main focus of future work.
6. References
[1] B. Haverkort, “Performance of Computer
Communication Systems, A Model-Based Approach”, John
Wiley & Sons, Ltd., 1998.
[2] D. Corne, M. Dorigo, and F. Glover (Eds.), “New ideas
in optimization”, Maidenhead, UK: McGraw-Hill, 1999.
[3] G. Di Caro, and M. Dorigo, “AntNet: A mobile agents
approach to adaptive routing”, Technical report TR-97-12.
Université Libre de Bruxelles, IRIDIA, 1997.
[4] G. Di Caro, and M. Dorigo, “AntNet: Distributed
stigmergetic control for communications networks”, In
Journal of Artificial Intelligence Research (JAIR) 9, 1998,
pp. 317–365.
[5] I. Chlamtac, M. Conti, and J. Liu, “Mobile ad hoc
networking: imperatives and challenges”, Ad Hoc
Networks, No. 1, 2003.
[6] M. Elizabeth, and T. Chai-Keong, “A Review of
Current Routing Protocols for Ad Hoc Mobile Wireless
Networks”, IEEE Personal Communications, 1999.
[7] M. Günes, U. Sorges, and I. Bouazizi,“ARA-The Ant-
Colony Based Routing Algorithm for MANETs”,
International Conference on Parallel Processing
Workshops (ICPPW’02), IEEE Computer Society Press,
2002, pp.79-85.
[8] M. Heissenbüttel, and T. Braun, “Ants-Based Routing
in Large Scale Mobile Ad-Hoc Networks”,
Kommunikation in Verteilten Systemen (KiVS), 2003.
[9] R. Schoonderwoerd, O. Holland, J. Bruten, and L.
Rothkrantz, “Ants for Load Balancing in Tele-
communications Networks”, Adaptive Behavior, Vol. 5,
No. 2, 1997, pp. 169-207.
[10] T. Camp, J. Boleng, and V. Davies, “A Survey of
Mobility Models for Ad Hoc Network Research”, In
Wireless Communication & Mobile Computing (WCMC),
Vol. 2, No. 5, 2002, pp. 483-502.
Proceedings of the International Conference on Information Technology: Coding and Computing (ITCC’05)
0-7695-2315-3/05 $ 20.00 IEEE