You are on page 1of 6

1

Ant Colony Optimization for


Mobile Ad-hoc Networks
Ajay C Solai Jawahar
Department of Electrical Engineering, Rutgers
ajaychak@eden.rutgers.edu

3) Although no guarantee of service can be provided, the


Abstract—The rapidly changing and unpredictable nature of protocol must be able to maximize the reliability of
Mobile Ad-hoc NETworks (MANETs) pose a wide range of packet in the network for the given conditions.
challenges like efficient routing, load distribution, congestion
avoidance, energy consumption, etc. Though a number of With these factors in mind, the key parameters to
routing protocols exists, which aim to provide effective routing
face while designing a routing protocol are
in MANETs, very few address the need or provide a plausible
solution for overall network management through routing in
1) Effective Routing
MANETs. This is the foremost requirement of the protocol – to
This project attempts to explore the phenomenon of successful discover and deliver the packet from the
emergence found in natural environments like harvester ants source to the destination. Some of the measures for
and aims to develop a similar routing protocol in manets with effective routing include Packet Delivery Ratio,
simple node-level management behavior resulting in overall percentage of Optimal Routes taken and average end-to-
system optimization. The proposed algorithm provides a end delay.
probabilistic multi-path routing algorithm and incorporates 2) Congestion Avoidance
path pheromones scents which constantly update the goodness of Strongly coupled with the previous parameter, this is
choosing a particular path based on congestion measure in
to ensure that the routing protocol does not congest a
addition to shortest-path metrics.
particular route/node thereby leading to packet drops or
.
even failure of the nodes.
Index Terms—ACO Routing, Emergence, Mobile Ad-hoc 3) Energy Consumption
Networks Most of the mobile nodes are laptops, pdas and other
portable devices which have a strong requirement for
energy consumption. Hence the routing algorithm must
I. INTRODUCTION minimize the energy consumption of the individual
nodes.
M OBILE Ad-hoc Networks have grown in popularity
over the recent advancements in technology as they
4) Load Balancing
Some of the nodes may be strategically located
possess the ability to provide instant wireless networking resulting in being present in most of the optimal routes of
solution in places/situations where no pre-deployed communication. Such nodes get unfairly overloaded
infrastructure exists. Manets prove to be an attractive solution leading to network congestion. Hence there is a need for
in a wide of environments ranging from scattered military balancing the loads on the individual nodes, though
deployment to simple network of a group of notebooks in an compromising on the optimal routes, thereby resulting on
office meeting. In all cases, there is a need for a group of fair load distribution.
nodes (laptops, desktops, pdas, cell phones, etc) to group Network Management of such a mobile network is hence a
together and create a network which can offer services like very big challenge. Also the fast changing nature and the ad-
file-sharing, messaging, resource sharing, etc. hoc necessity of the network prevents the choice of a
Hence the primary goal in a mobile network is to efficiently Centralized Solution which can decide the best route to route
establish one or more routes between two nodes so that they packets and at the same time minimize the different
can communicate reliably. Such a network is characterized by parameters like congestion, load, etc. Also no single node can
the following challenges. take up the job of centralized manager due to the limited
1) The network topology can change dynamically due to energy and processing capabilities of mobile nodes. Hence
the random movement of nodes. this has resulted in the need for a distributed approach which
2) Also any node may leave/join the network and the involves limited processing and power from the individual
protocol must adapt accordingly. nodes but which work towards a concerted goal of routing
and network management.
Different routing protocols have been developed for
Manets, some proactive like DSDV and some on-demand like
AODV and DSR. However most of these protocols only
2

attempt to solve one challenge – Effective Routing – and the scents of pheromones on the path. In this way, a number
provide algorithms for minimal end-to-end delay, maximum of paths might exist from the nest to the food source. Also the
throughput, etc. shortest path will be the one with the highest pheromone
Hence there is a need to develop a unified routing protocol, scent (due to shorter round-trip time) (see Fig 1) and also
which also results in overall network management in addition naturally the path with the highest concentration of ants.
to just packet routing. In this attempt, this project looks at the At the same time, it must be noted that usually multiple
characteristics of Emergence behavior exhibited in natural trails exist from the nest to the food source. When a
environments by studying the emergence behavior in ants and previously short route get blocked/lengthened due to an
presents a unified solution for network management through
obstacle en route, the alternate short route get strengthened
routing.
with higher pheromone content due to shorter end-to-end
travel time and more ants move to this route. Hence the path
II. EMERGENCE IN HARVESTER ANTS can also dynamically adapt to fast changes in the
environment.
The phenomenon of emergence found in natural systems
show how simple behavioral patterns from participants give
rise to complex self-regulatory behavior of the entire III. ROUTING ALGORITHM
system[1]. The swift co-ordinated flight of a group of birds
with very little visual/vocal communication and the concerted The Routing Algorithm proposed is an on-demand reactive
effort of an ant colony in gathering food, building nests, etc algorithm similar to AODV/DSR but has been built to take
are some of the vivid examples of emergence in natural into consideration all the challenges in Manets, thereby
world. providing active network management into addition to simple
A colony of harvester ants have a wide range of duties routing.
like collecting food, building/guarding the nest, removing the This section describes in detail the proposed algorithm and
dead ants, etc and have very simple one-to-one how route discovery and route maintenance is performed.
communication. Here the individual messages passed between The current implementation of the project tries to optimize
ants are very insignificant, but the collective messages help in the following factors
coordinated work control of ants without the presence of a - Efficient Routing (measured by throughput and
centralized control system. optimal routes)
For communication between each other, ants use scents - Congestion Avoidance (measured by Node Level
called pheromones. They can produce few different types of Congestion Measurement)
pheromones – usually one each to signify different work - Load Balancing (caused by multi-path routing)
categories like collecting food trails, signifying emergency, The current proposal does not do explicit energy
management in the nodes but ensures fair energy distribution
by using a balanced-load algorithm.

A. Node Level Congestion Measurement


Measuring the congestion level at a particular node has
always been a difficult task. A number of approaches have
been taken and different measures have been considered to
quantize congestion. In this project, the scheme for
measurement of Node Congestion Metric is as discussed in
the paper “Accurate and Energy-efficient Congestion Level
Measurement in Ad-Hoc Networks” [2].
Fig. 1. Route Discovery Phase, shows the propagation of Forward ANTS from Every node calculates a Node Congestion Metric which
source to destination. Also shows the calculation of Route Efficiency Metric.
Source: Internet gives an indication on the amount of congestion at the node.
This metric varies between 0 (Very High Congestion) to 1
moving dead ants, etc.
(Least Congestion). The Node Congestion Metric is
Let us consider the example of food collection. Initially the
calculated based on the following factors
ants spread out in all directions in search of food. When an
• Buffer Occupancy
ant finds a food source, it collects the food and on returning
• Channel Load Measurement
back, marks the trail with Pheromones. These pheromones
are dropped at regular intervals to act as a trail. Also the • Packet Drop Rate
pheromones slowly disappear over time. So they act as a
guiding trail to other ants which begin to follow this path. In
the same way, ants which trace a particular path strengthen
3

B. Route Discovery Also similarly in the case of harvester ants, the shortest path
When a source needs to send a packet to a particular node, (which has the highest pheromone scent) serves as the best
it first checks the cache for existing routes. When no routes path for the ants.
are known, it broadcasts Forward Request ANTS which are However research has shown that in MANETs , the best
propagated through the network till it reaches the destination. path for routing (considering the overall network benefit as
This process can be compared to ants initially spreading out well as node benefit) is not necessarily the shortest path but
in all directions in search of food source. instead the path which optimizes number of hops (length of
When a packet reaches the destination, the destination path), congestion along path and load balancing.
node sends a Backward ANT for every Forward Request Hence building on the existing Node Congestion Metric,
this algorithm calculates Route Efficiency Metric (REM)
based on the Individual Node Congestion values.

C. Calculation of Route-Efficiency Metric


Whenever a node sends a packet (Forward Request ANT,
Backward ANT, normal data packet, etc), the packet has a
metric value called Route Efficiency Metric (REM). Initially
the source who originates the packet sets it to 1 (Minimum
Congestion), although the Node Congestion value of source
may be less than 1.
Fig. 1. Route Discovery Phase, shows the propagation of Forward ANTS from When the packet is being forwarded/broadcasted, all the
source to destination. Also shows the calculation of Route Efficiency Metric.
intermediate nodes multiply the REM with the individual
ANTs that it receives. Thus when different Forward ANTs Node Congestion Metrics. Then the REM can be defined as
reach the destination through different routes, the destination
sends a Backward ANT for each of them. This is to ensure Route Efficiency Metric = ∏ (Node Congestion values of
that multiple paths exist between the source and the intermediate nodes from source to destination)
destination. Synonymously, in the case of ants, initially
multiple paths exist between the nest and the food source. Fig 2. shows the calculation of Route Efficiency Metric
Slowly the best path (which for ants is the shortest path) gets during the propagation of Forward ANTS and Fig 3. shows
strengthened through increased pheromone count. the Route Efficiency Metric being piggybacked on the
Backward ANTS from the destination to the source.
How does this work? This calculation is based on the
The Backward ANTs from the destination reach the source assumption that the individual node congestion measure is
nothing but a probability that a node can forward a packet
successfully to the next hop. Hence the probability of sending
a packet successfully from a particular source to the
destination is nothing but the product of the individual
probability measures.
However this does arise a question that since Node
Congestion Metric is considered as a linear measure, the
results may not be correct as high values of Node Congestion
(say 0.6 or 0.8) have the same impact on the overall
congestion as low values (like 0.1 or 0.05). Hence the
algorithm converts the linear weights of Node Congestion
Metric to a non-linear weight using the following equation
Fig. 1. Route Discovery Phase, shows the propagation of Backward ANTS
from destination to source New Node Congestion = (1 – (1- Node Congestion)3 )
and the source stores the route in the cache. Hence this way,
multiple routes to the destination will be stored in the source. The following diagram show the conversion of Node
Congestion value from linear to a non-linear scale so that
Next the source needs to decide the best route to follow. In higher values of Node Congestion have lesser impact and as
most conventional techniques like DSR/AODV, the source the congestion decreases, it has a much greater impact.
chooses the shortest path to the destination with the
assumption that the same helps in reducing energy and time.
4

Probability of Route
= Pheromone Count of Route / ∑ (Pheromone count
of all routes to destination)

Once the probabilities are assigned to the individual routes,


the node then chooses a route as per this probability. Hence
good routes (based on length and congestion) have higher
probability of being chosen while all alternate routes are still
kept fresh.

E. Route Maintenance
Fig. 1. Route Discovery Phase, shows the propagation of Backward ANTS Route Maintenance plays a very important role in manets
from destination to source as the network keeps dynamically changing and routes found
good during discovery may turn to be bad due to congestion,
Hence when a destination node receives a Forward
signal strength, etc.
Request ANT, it takes the REM values and piggybacks the
Hence when a node starts sending packets to the
same with the Backward ANT. Hence a source node receives
destination using the Probabilistic Route Finding algorithm
not only the route, but also the REM value for that route. The
explained above, it is essential to find the goodness of a route
REM value for a route is stored in the cache along with the
regularly and update the pheromone counts for the different
route.
routes at the source nodes. To accomplish this, when a
destination node receives a packet, it probabilistically sends a
D. Probabilistic Route Choosing Algorithm Congestion Update message to the source which informs the
Usually when a source wants to find a route to the source of the REM value for that route. This Congestion
destination, it chooses the route which is the best path Update message also serves an ACK to the source. These
(usually shortest path or minimal congestion path, etc). Congestion Update messages are sent probabilistically (say
However studying ant behavior, although the best path with prob of 0.1), so that they have minimum overhead and
(shortest path) has the maximum pheromone concentration do not increase much the congestion if the route is already
and hence maximum ants in the path, there always are small overloaded.
trails of other non-efficient routes which exist between the On receiving the Congestion Update message from the
nest and the food source. destination, the source updates the cache with the REM
A similar technique will be of great use in MANETs, more values as shown below
so because of the ever-changing nature of the network and
hence to need to keep other routes active for contingency in New REM value = α * (Old REM Value) + (1- α ) *
case the best route gets overloaded or goes down due to a (Received REM value)
faulty node.
Hence this algorithm uses a Probabilistic Route Choosing The above is done so that a particular fluctuation in a Node
Algorithm in which initially the pheromone scents is which may result in abnormal Node Congestion and hence
calculated for every route (to the particular destination) and REM value does not affect the overall REM value for that
then probabilistically a route is chosen based on the route.
Pheromone values. Initially the pheromone count is Also this makes sure that the pheromone counts for the
calculated as below using the REM value and the length of active routes are constantly being refreshed and re-
the route. This is done so as to reach a compromise between strengthened so that better choices are made which adapt to
the shortest route and the least congested route. the ever dynamic nature of the mobile networks. Hence the
probability of a route being selected constantly changes as the
Pheromone Count of Route = (REM)L REM value for the route changes. Further work is suggested
where REM = Route Efficiency Metric in the future to incorporate the Energy Consumption values
L = Length of the Route for the individual routes so that the choice also optimizes the
same.
Similarly the pheromone count is calculated for all the The suggested algorithm strives to reduce packet loss, as it
routes. Next probabilities are assigned to the individual routes keeps track of queue length, channel load, buffer, etc for all
based on their pheromone values. the nodes en route and automatically reduces traffic through a
route on sensing congestion and uses alternate routes.
5

F. Comparison with Existing ACO Algorithms All the nodes in the network are constantly moving and
Ant Colony Optimization algorithms and their applications have a random speed selected in the range of 0-5m/sec.
in Mobile Ad-hoc Networks have been studied for long. A The key performance metrics evaluated in the simulations
couple of popular algorithms which aim to alleviate the are
problems in Mobile Ad-hoc Networks are the AntHocNet [3] • Packet Delivery Ratio (Ratio of number of packets
delivered to number of packets sent)
Algorithm and ARA (Ant colony-based Routing Algorithm)
• Average Hop / Packet (Ratio of number of individual
[4].
packets to number of messages sent)
Both the routing algorithms extensively use the concept of
• Routing Overhead (Ratio of number of control packets to
pheromones and pheromone tables. AntHocNet, a more
number of messages sent)
lucrative solution of the two, is a hybrid algorithm, which has
• Energy , Load Distribution (Varied in direct relation to
both proactive and reactive features. Hybrid Algorithms, the number of packets forwarded/sent by the individual
despite their better performance are not very popular due to nodes)
the heavy overhead of the proactive algorithm (continuous
exchange of routing information, etc between neighbors).
Also AntHocNet uses end-to-end delay as a metric to B. Simulation Results
calculate congestion at a node, which may not yield accurate The first set of simulations is performed on a 10-node
results as end-to-end is affected by both congestion as well as network in a 400x400m2 area. The results are the average of
the length of the route from source to destination. the performance metrics over 3 simulations with different 10-
ARA is an on-demand routing algorithm similar to DSR, node scenarios with the same traffic. The traffic load here are
AODV and uses pheromone tables at each node to calculate 2 sources sending 1024 bytes data at an interval of 0.1
seconds.
the best route to reach the destination. However the
The following table shows the comparison of the
pheromones merely enhance routes without any dependency
performance of the proposed algorithm with the DSR routing
on congestion, load or energy metrics. Such a mechanism algorithm.
always enhance the shortest routes although it might be
TABLE I
congested or non-efficient in terms of energy/load
10-NODES, 2 TRAFFIC SOURCES
distribution.
Hence the suggested algorithm, being on-demand, has low Metric DSR ANT Algorithm
overhead and at the same time tries to optimize all the Packet Delivery 99.962 99.936
parameters (energy, congestion, load balancing) of the Ratio
Average Hop / 2.388 2.402
network along-with providing efficient routing. Also the Packet
congestion measurement algorithm suggested helps in Overhead
accurate measurement of congestion at a node based on Control Packets 17 448
Message Packets 7762 7836
multiple factors as buffer length, drop ratio and channel Routing Overhead 0.22% 5.72%
occupancy.

From Table 1, we see that the performance of both ANT


IV. SIMULATION RESULTS Algorithm and DSR are similar in terms of packet delivery
The proposed algorithm has been evaluated though a ratio and average hops. However the ANT algorithm has
number of simulation tests and compared with DSR[8] On- higher overhead due to the continuous exchange of Backward
demand routing algorithm. ANTS in the route maintenance phase for updating the
goodness of a particular route.
A. Simulation Environment
However the biggest difference is in load balancing and
The simulation of the algorithm was carried out using ns-2 energy management. Fig 4 shows the number of packets
(version 2.28) simulator [6],[7]. Two different scenarios are
sent/forwarded by each node in the network. The DSR results
considered in this case, one with 10 nodes in a 400x400m2
show a chaotic distribution of packets amongst the nodes.
network and the other with 50 nodes in a 670x670m2
network. All simulations are carried out for 250 seconds. The Some of the nodes (nodes 8 & 9) forward an extremely high
sources in the network are CBR (constant bit rate) traffic number of packets results in high energy utilization while
generators. The data packets used for simulations are 1024 others are relatively idle. However in the Ant Algorithm,
bytes. since multiple routes are used to send packets, the loads are
The MAC layer uses the 802.11b protocol for the wireless even distributed amongst most of the nodes. Notice that some
standard. The radios use Two Ray Ground propagation model nodes have very low packets sent/forwarded as they do not lie
and have a receiving range of 250 metres. in the route from source to destination.
6

Packet Flow Analysis Also some nodes handle heavy traffic forwarding (more than
18000
15000 packets). However in Ant Algorithm, more nodes
16000
(about 22%) are involved in active traffic forwarding and also
14000
the traffic is more distributed amongst the nodes. Thus Ant
Number of Packets sent/forwarded

Algorithm provides better load and energy balancing than


12000
DSR.
10000
DSR
ANT-DSR
8000

6000
V. CONCLUSION & FUTURE WORK
4000
Mobile Ad-hoc Networks provide a wide array of
2000
challenges in routing and network management due to their
0
0 1 2 3 4 5 6 7 8 9
dynamic and distributed nature and various protocols have
Node Number been studied and implemented in view of these needs and
Fig. 4. Load Distribution in a 10-Node Scenario challenges. In this paper, an alternate approach inspired by
concepts of emergence and self-organization in biological
systems, has been discussed and implemented.
The second set of simulations is performed on a 50-node
The approach tries to optimize routing by finding the best
network in 670x670m2 field area. Also 3 cbr traffic sources
shortest path, optimize congestion by incorporating
are used which send packets of size 1024 bytes at an interval
congestion metrics in route discovery and maintenance and
of 0.1 seconds.
also optimize energy and load balancing by using the
TABLE 2 probabilistic route choosing algorithm. Also the comparison
50-NODES, 3 TRAFFIC SOURCES
of this algorithm with de-facto DSR algorithm has been
Metric DSR ANT Algorithm presented.
Packet Delivery 94.92 99.3
This algorithm has a lot of scope for future improvements.
Ratio Though the algorithm has better control over congestion and
Average Hop / 2.756 3.122 load, its adaptability to dynamic node movements is not very
Packet
Overhead different from existing algorithms. Hence future work lies in
Control Packets 122 1421 incorporating factors like signal strength into the route
Message Packets 18726 18850 metrics so as to predict link breaks before they actually occur
Routing Overhead 0.65% 7.54%
and redirect to other routes. Also extensive testing needs to
be performed using the current algorithm under varying
traffic and movement scenarios.
From Table 2, we see that although the ANT algorithm has
a higher routing overhead than DSR, it has a better
performance in terms of packet delivery ratio. Also the REFERENCES
increase in average hop/packet is around 0.4 hops.
[1] G. Serugendo, N. Foukia, S. Hassas, A. Karageorgos, S. K. Mostefaoui, O.
Also from Fig 5, we see that the load distribution is uneven F. Rana, M. Ulieru, P. Valckenaers, C. Van Aart, "Self-organisation:
in DSR with only 10% of the nodes involved in active traffic. Paradigms and applications", Engineering Self-Organising Systems, Vol.
2977, 2004
Packet Flow Analysis
[2] Jaewon Kang, Yanyong Zhang, Badri Nath, “Accurate and Energy-
efficient Congestion Level measurement in Ad-hoc Networks”
45000
[3] Frederick Ducatelle, Gianni Di Caro and Luca Maria Gambardella,
40000
“Using ANT Agents to combine reactive and proactive strategies for
routing in mobile ad-hoc networks”
35000
[4] Mesut Gunes, Udo Sorges, Imed Bouazizi, “ARA – The Ant-colony based
Number of packets sent/forwarded

30000
routing algorithm for MANETs”, International workshop on Ad-hoc
Networking (IWAHN 2002)
25000
DSR [5] Apostolos Malatras, George Pavlou, Stylianos Gouveris,
20000
Ant-DSR Sivapathalingham Sivavakeesar, “Self configuring and optimizing Mobile
Ad-hoc Networks”,Proceedings of the Second International Conference on
15000
Autonomic Computing.
10000
[6] http://www.isi.edu/nsnam/ns/
[7] The NS Manual , VINT Project (Editors: Kevin Fall and Kannan Vardhan)
5000 [8] David B Johnson, David A Maltz, Yin-chun hu, “The Dynamic Source
0
Routing Protocol for Mobile Ad-hoc Networks” , IETF Draft.
2 3 4 5 10 15 16 27 44 47 [9] Eitan Altman and Tania Jimenez, “NS Simulator for beginners”
Node Number

Fig. 5. Load Distribution in 50-node scenario. Note: Only nodes with more
than 500 packets forwards are plotted in the graph.

You might also like