4 Multiple Feasible Paths in Ant Colony Algorithm

8/6/2019 4 Multiple Feasible Paths in Ant Colony Algorithm

1/7

2011 Shweta Modi, Jitendra Prithviraj. This is a research/review paper, distributed under the terms of the Creative Commons

Attribution-Noncommercial 3.0 Unported License http://creativecommons.org/licenses/by-nc/3.0/), permitting all non-commercial use,distribution, and reproduction inany medium, provided the original work is properly cited.

Global Journal of Computer Science and TechnologyVolume 11 Issue 4 Version 1.0 March 2011

Type: Double Blind Peer Reviewed International Research Journal

Publisher: Global Journals Inc. (USA)

Online ISSN: 0975-4172 & Print ISSN: 0975-4350

Multiple Feasible Paths in Ant Colony Algorithm for mobile Ad-

hoc Networks with Minimum OverheadBy Shweta Modi, Jitendra Prithviraj

Abstract-Mobile ad-hoc networks are infrastructure-less networks consisting of wireless, possibly

mobile nodes which are organized in peer-to-peer and autonomous fashion. The highly dynamic

topology, limited bandwidth availability and energy constraints make the routing problem achallenging one. Ant colony optimization (ACO) is a population based meta-heuristic for

combinatorial optimization problems such as communication network routing problem. In real life,

ants drop some kind of chemical substances to mark the path that they used. Then on their way,

back they choose the path with the highest pheromones which becomes the shortest path. But Ant

net Algorithms may cause the network congestion and stagnation. Here, multiple optimal paths are

proposed with negligible overhead in spite of single optimal path in Ant net routing algorithm,

so that the problem of stagnation can be rectified. This paper proposes an improved Multiple

Feasible Paths in Ant Colony Algorithm for mobile Ad-hoc Networks with Minimum Overhead.

Keywords:MANET, Routing, Antnet, agent, swarm Intelligence, Adaptive, Ant colony optimization,

mobile agent,DSR,OLSR,ADODV

Classification:GJCST Classification: C.2.2, C.2.1

MultipleFeasiblePathsinAntColonyAlgorithmformobileAd-hocNetworkswithMinimumOverhead

Strictly as per the compliance and regulations of:


2/7

2011 Global Journals Inc. (US)

Multiple Feasible Paths in Ant Colony Algorithmfor mobile Ad-hoc Networks with Minimum

Overhead

Miss Shweta Modi1, Mr. Jitendra Prithviraj2

Abstract- Mobile ad-hoc networks are infrastructure-lessnetworks consisting of wireless, possibly mobile nodes whichare organized in peer-to-peer and autonomous fashion. Thehighly dynamic topology, limited bandwidth availability andenergy constraints make the routing problem a challengingone. Ant colony optimization (ACO) is a population basedmeta-heuristic for combinatorial optimization problems such ascommunication network routing problem. In real life, ants dropsome kind of chemical substances to mark the path that theyused. Then on their way, back they choose the path with thehighest pheromones which becomes the shortest path. But Antnet Algorithms may cause the network congestion andstagnation. Here, multiple optimal paths are proposed withnegligible overhead in spite of single optimal path in Antnet routing algorithm, so that the problem of stagnationcan be rectified. This paper proposes an improved MultipleFeasible Paths in Ant Colony Algorithm for mobile Ad-hocNetworks with Minimum Overhead.Keywords: MANET, Routing, Antnet, agent, swarmIntelligence, Adaptive, Ant colony optimization, mobileagent,DSR,OLSR,ADODV

I. INTRODUCTIONoday with the fast growth of Internet, everybodywants to get connected to the internet. Millions ofpeople use the Internet for daily business all over

the world. Today the Internet has become very large,complex and dynamic. Failures and challenges occur atevery step because of traffic flow from one part ofnetwork to another part. Routing is the process ofselecting paths in a network to send traffic. Routing is animportant aspect of network communication, whichaffects the performance of any network, since othercharacteristics of the network like throughput; reliabilityand congestion depend directly on it. In packet

switching networks when a packet travels from asource to destination, it has to pass through anumber of networks with varying characteristics. So anideal routing algorithm is one which is able to deliver thepacket to its destination with minimum amount of delay.It must be adaptive and intelligent enough to make thedecisions. The growing size and increasing demands ofInternet impelled the study of more powerful routing

Author1-Dept. of CSE, SRIT Jabalpur, IndiaE-mail- [email protected], ME (CSE), SRIT Jabalpur, IndiaE-mail- [email protected]

algorithms, which can optimize the flow of traffic.The routing algorithms currently in use (e.g.

OSPF, RIP, DSR,ADODV and BGP) are not sufficient totackle the increasing complexity of such networks. Theyare not adaptive, intelligent and fault intolerant. Therouting tables in them are updated by exchangingrouting information between the routers. Different routingprotocols use different approaches to exchange therouting information. There are mainly two approaches for

routing algorithms, distance vector algorithms and link-state algorithms. Distance vector algorithms use theBellman algorithm. This approach assigns a number, thecost, to each of the links between each node in thenetwork. In link-state algorithms for example, OpenShortest Path First (OSPF), the routers exchange link-state information by flooding the link state packets. Thelink state updates are generated only when the linkstatus changes. Once a node has obtained topologyinformation of the entire network, Dijkstras algorithmis generally used to compute the shortest path. The twomain performance metrics that are affected by the

routing algorithm are throughput and average packetdelay. Nodes and links can fail, and congestion canarise in some areas. Thus, the routing algorithm needsto modify its routes, redirecting traffic and updatingdatabases very quickly and adaptively.

In recent years, a new kind of routing protocolsinfluenced by software agents called Ant Based routingis developed. S. Appleby and S. Steward were the firstones to introduce the concept of software agentsused for control in telecommunication networks .Theresearch process continued and it was applied toconnection oriented networks [3]. Ant based routingwas then applied to packet based connection lesssystems [4].This agent based approach was furtherresearched and was modified for adaptive routing[5].Swarm intelligence provides a promising alternativeto traditional routing algorithms by utilizing mobilesoftware agents for network management. Although, anant [1], [2] is a simple and unsophisticated creature,collectively a colony of ants can perform useful taskssuch as building nests, and foraging (searching for food)[1],[2],[3].What is interesting is that ants are able todiscover the shortest path to a food source and to sharethat information with other ants through stigmergy [1]-[5]. Stigmergy is a form of indirect communication used

T

G

lobal

JournalofComputerScience

andTechnology

VolumeXIIssueIV

VersionI

23

M

arc

h2

011


3/7

Multiple Feasible Paths in Ant Colony Algorithm for mobile Ad-hoc Networks with Minimum Overhead


by ants in nature to coordinate their problem-solving activities. Ants achieve stigmergiccommunication by laying a chemical substance calledpheromone.

II. ANT NET COLONYOPTIMIZATIONACO [2] is a met heuristic in which a colony of

artificial ants cooperates in finding good solutions todiscrete optimization problems. Each ant of the colonyexploits the problem graph to search for optimalsolutions. An artificial ant, unlike natural counterparts,has a memory in which it can store information about thepath it follows. Every ant has a start state and one ormore terminating conditions. The next move is selectedby a probabilistic decision rule that is a function oflocally available pheromone trails, heuristic values aswell as the ant's memory. Ant can update thepheromone trail associated with the link it follows. Onceit has built a solution, it can retrace the same pathbackward and update the pheromone trails. ACOalgorithm is interplay of three procedures as describedin [2].

1. Construct ant solutions: This proceduremanages a colony of ants that concurrently andasynchronously visit adjacent states of theconsidered problem by moving throughneighboring nodes of the solution space of theproblem's construction graph.

2. Update pheromones: It is the process by whichpheromone trails are modified. The trail valuecan either increase, as ants deposit pheromoneon the components or connections they use, or

decrease, due to pheromone evaporation. Netincrease/decrease in pheromone value at agiven location on trail is determined bydifference of deposition and evaporation.

3. Daemon actions: This procedure is used toimplement centralized actions which cannot beperformed by single ants.

III. ANT NET DATASTRUCTURES AntNet is an ACO algorithm for data network

routing proposed by Gianni Di Caro and Marco Dorigo[2]. Mobile agents (artificial ants) act concurrently and

independently, and communicate in an indirect way(stigmergically), through the pheromones they read andwrite locally on the nodes. Each network node k storestwo data structures:

1) Routing table TkFor each possible destination d and for each

neighbor node n, Tk stores a probability value Pndexpressing the goodness of choosing n as next nodewhen the destination node is d:

Pnd = 1, d [1, N], N k = neighbors (k)nNk

Probability value Pnd represents the pheromoneconcentration along the link from node k to neighbornode n for destination node d.

2) Traffic model MkMk (d, d

2,Wd) is a statistical model of the trafficsituation over the network as seen by node k. It is

described by the sample mean d and the variance d2

computed over the trip times experienced by the artificialants, and by a moving observation window Wd used tostore the best value W best_d of the artificial ants trip time.

IV. ANT NET ROUTING ALGORITHMAntNet algorithm, as proposed by Di Caro and

Dorigo, is as follows1. At regular intervals t from every network node

s, a forward ant Fsd is launched toward adestination d to discover a feasible, low-costpath to that node and to investigate the loadstatus of the network along the path. If fsd is ameasure (in bits or in number of packets) of thedata flow sd, then the probability of creatingat node s a forward ant with node d asdestination is

Psd = (fsd)

N

fsii=1

While traveling toward their destination nodes,the forward ants keep memory of their paths and of the

traffic conditions found. The identifier of every visitednode i and the time elapsed since the launching time toarrive at this i-th node are stored in a memory stack.

2. At each node i, each forward ant headedtoward a destination d selects the node j tomove to, with a probability Pijd computed as

normalized sum of the pheromone ijd with aheuristic value ij taking into account the lengthof the j-th link queue of the current node i :

P ijd = ijd+ ij

1+ (| N i| -1)

The heuristic value ij is a normalized valuefunction of the length qij of the queue on the linkconnecting the node i with its neighbor j:

ij = 1- (q ij)|Ni| qi1i=1The value of weighs the importance of the

heuristic value with respect to the pheromone values.

3. If a cycle is detected, the cycle's nodes areremoved and all the memory about them isdeleted. When an ant reaches a node that isalready in its memory, a cycle is detected and

G

lobal


andTechnology

VolumeXIIssueIV

VersionI

24

M

arc

h2

011


4/7


all the nodes until this recurrent node aredeleted from the ants memory.

4. When the destination node d is reached, theagent Fsd generates backward ant Bds,transfers to it all of its memory, and is deleted.

5. The backward ant takes the same path as thatof its corresponding forward ant, but in theopposite direction.

6. Arriving at a node i coming from a neighbornode, the backward ant updates the local modelof the traffic Mi and the pheromone matrix Ti, forall the entries corresponding to the destinationnode d.

1) Update traffic model MiThe estimated mean and variance are updated asfollows:

id id+ (oid - id)

2

id2

id+ { (oid - id)2

- 2

id}where oid is the observed ant's trip time from node i todestination d. The factor weighs the number of mostrecent samples that will really affect the average

2) Update pheromone matrix TiThe backward ant Bds moving from node f to node i

increases the pheromone values ifdifdifd+ r

. (1 - ifd)

Pheromones ifd for destination d' of the otherneighboring nodes j, j Ni, j f, evaporate implicitly bynormalization.

ifd

ifd

+ r. (ifd

) ,j Ni, j f,

V. LIMITATIONS Although Experiments of AntNet have shown

very promising results, antnet has outperformed underdifferent experimental conditions with respect to otherdynamic routing algorithms e.g. RIP, OSPF. Still thereare some problems with this adaptive algorithm. One ofthe major problems is that the network gets trappedbecause a node prefers a link with higher probability to adestination when choosing an outgoing link say no tosend a packet. If link no keeps good condition for a longtime, its probability to that destination will be very high.Such a condition may cause congestion of no; it alsoreduces probability of selecting other paths. Hence thenode will stuck to this outgoing link and loose itsadaptive ability. This is called the problem ofstagnation. Stagnation is reached when a nodereaches its convergence. Stagnation is a very criticalbecause

1. no may lose its optimality if it gets congested2. If the network gets fails at any time then the

most preferred path no may becomeunavailable

3. Other non-optimal paths may become optimaldue to changes in network topologyMany researchers have tried to provide the

solution for the same. These methods are Evaporation Aging Limiting and Smoothing Pheromone

But all the methods are very complex andneeded extra overhead. Here, multiple optimal paths areproposed so that the congestion and stagnation oversingle optimal path can be avoided.

VI. PROPOSED LOGIC According to original Antnet Algorithm ,Routing

Table is a local data-base that helps router todecide where to forward data packets. It contains theinformation which specifies the next (neighbor) node thatshould be taken by a packet to get to any possibledestination in the network. The columns of the table

represent the neighbor nodes and rows represent thepossible destinations.

The column wise values in the table are pickedup and a sorting algorithm is executed onthese values.

The sorted values ranging from higher to lowerare stored in a temporary array.

The difference d amongst the adjacent value iscalculated and is compared to some thresholdvalue say pm.

If the difference d is less than pm then thosevalues are selected and comparison amongstthe adjacent values is continued until difference

is greater than pm. Otherwise at the very first occurrence of

difference greater than pm, the comparison isstopped and the corresponding value(s) in thearray is (are) selected.

The interfaces corresponding to these valuesare stored in a new routing table which will havethe same structure as the original one, butobviously, the new table will have less numberof rows.

VII. A NETWORKSIMULATORTOOL(Ns2)

We chose ns-2 for the implementation of AntNetbecause it is an event driven simulator which enablesthe generation and forwarding of packets according tothe route logic. Also trace files provide nice details ofpacket flow that enables to check the correctness andaccuracy of algorithm.

VIII. ANT NET IMPLEMENTATIONWe could successfully extend network simulator

ns-2 to implement AntNet simulation. Ant packets of twotypes representing forward ants and backward ants flow


G

lobal


andTechnology

VolumeXIIssueIV

VersionI

25

M

arc

h2

011


5/7


through the network according to the algorithm andkeep a memory of the path traversed to discover optimalsolution. We defined a routing agent called Antnet agentwhich is responsible to implement route logic andperform the protocols functionality by processing antpackets according to the flow of the algorithm. Routingagent is implemented through class Antnet derived from

class Agent. We defined Tcl hooks that enablesimulation of AntNet routing algorithm through Tcl script.Hence, it can be simulated for varying simulation timeand time interval at which ants are launched.

1) Ant PacketsWe have implemented artificial ants by defining

a new packet type in ns-2. Besides information onpacket source, destination, sequence number andlength, the packet header also has a field that identifiesit as forward ant or backward ant. A local data structurerepresents the memory of a packet, which stores nodeaddresses and trip time to corresponding nodes that the

packet traversed.

2) Antnet Routing AgentWe have defined class Antnet inherited from

class Agent. This implements the Antnet routing agent towhich a node can be associated. The routing table andlocal traffic model at a node are stored as local datastructures in the Antnet agent. It is responsible forcreating and forwarding forward ant packets andbackward ant packets. The modules in the Antnet classimplement the functionality of receiving ant packets,processing them according to the flow of the algorithm

which includes to determine destination node and nexthop node, and forwarding them to Antnet agentassociated with next hop node. Whenever Antnet agentreceives a forward ant, it adds the node address and triptime to the ant's memory, determines next node to whichit has to be forwarded and sends it to Antnet agentassociated with that node. When Antnet agent receives aforward ant destined to itself, it creates a backward antpacket and sends it along the reverse path as stored inant's memory. When Antnet agent receives a backwardant packet, it updates the routing table and local trafficmodel and sends it further along the path that it hasbeen retracing. Upon complete traversal of path, thebackward ant packet is destroyed. The agentimplements cycle detection and elimination, packetforwarding and update of traffic model and routing tableaccording to the AntNet algorithm.

3) ResultsWe simulated AntNet on ns-2 for a number of

topologies. It was observed that AntNet constructsprobabilistic routing tables, wherein better paths amongthe available paths have higher pheromoneconcentration.

Now follows the description of our experimentalsimulation on ns-2. Simulation parameters are asfollows:

1. Simulation time = 11.1s. {This is the time whichthe algorithm takes to converge.}

Fig.1. Network Topology used for simulation results onns-2

2.

Time interval at which forward ants are launched= 0.02s

As an example for the explanation of results,consider a packet having source node 2 and destinationnode 1 in Fig.1.It can be observed that among all theneighbor nodes of node 2, node 0 provides the bestpath of only one hops for node 0.Nodes 3 provide next

best paths, both of them being almost equally good.Node 3 is the worst of all possible paths, because itresults in a cycle between nodes 1 and 2. Theprobability of selecting node 0 as the next hop is0.817340, which is the highest among all neighbors.Next higher values are those corresponding to node 3(0.040842) both of them being almost equal.

The simulation time of 10.8 seconds representsthe saturation time required for the convergence of thealgorithm with forward ant packets launched every 0.03seconds. Since, routing tables generate the best paths;the time taken by packet to travel from source todestination must be the minimum possible.

IX.

CONCLUSION

We successfully simulate AntNet on Network

simulator, ns-2. Our experiments generated correctrouting tables for all the topologies simulated. In thispaper an improved version of the AntNet algorithm isproposed for mobile ad-hoc network. In the improvedversion, more than one optimal outgoing interfaces areidentified as compared to only

one path, which are


G

lobal


andTechnology

VolumeXIIssueIV

VersionI

26

M

arc

h2

011

Simulation Result


6/7


supposed to provide higher throughput and will beable to explore new and better paths even if the networktopologies gets changed very frequently. This willdistribute the traffic of overloaded link to other preferredlinks. Hence the throughput of the network will beimproved and the problem of stagnation will be rectifiedin mobile ad-hoc network.

X. FUTURE WORKThis algorithm is efficient for end to end delay

and packets delivery ratio. We will simulate ourproposed algorithm with one of network simulators andwe compare the scheme performance with other routingalgorithms and simulator.

References Rfrences Referencias

1. S. Appleby and S. Steward, Mobile softwareagents

for control in telecommunicationnetworks, BT echnol. vol. 12, no. 2, 1994.

2. R Schoonderwoerd, O.Holland, J.Bruten, andL.Rothkrantz, Ants for Load Balancing inTelecommunication Networks, Hewlett PackardLab., Bristol, U.K., Tech. Rep. HPL-96-35, 1996.

3. E.Bonabeau, D.Suyers,Routing intelecommunication networks with smart likeagents, Proceedings of LATA, 1998.

4. D.Subbramanian, P.Druschel, and J.Chen,Antsand reinforcement learning: A case study inrouting in dynamic networks, Proces Alto,Morgan Kauffman, 1997, pp 832-838

5. M. Dorigo and G.D.Caro,Antnet: A mobileagents approach to adaptive routing, TechReport, University Libre de Bruxelles, IRIDIA,1997.

6. G.D.Caro and M.Dorigo,Antnet: Distributedstigmergetic control for communicationsnetworks, Journal of Artificial IntelligenceResearch, 9, 1998 vol.9, pp. 317-365, 1998.

7. G.D.Caro and M.Dorigo, Ant colonies foradaptive in packet-switched communicationsnetworks, in Proc. 5th Int. Conf. ParallelProblem Solving from Nature, Amsterdam, TheNetherlands, Sept 27-30, 1998

8.

G. D. Caro and M. Dorigo, Two ant colony

algorithms for best-effort routing in datagramnetworks, in Proc. 10th IASTED Int. Conf.Parallel Distributed Computing Systems, 1998,pp. 541-546.

9. M. Dorigo, G. D. Caro, and L. M. Gambardella,Ant algorithms for discrete optimization, Artif.Life, vol. 5, no. 2, pp. 137-172, 1999.

10.B. Baran and R. Sosa, A new approach for

AntNet routing, presented at the Proc.9th Int.Conf.Computer Communications Networks, LasVegas, NV, 2000.


G

lobal


andTechnology

VolumeXIIssueIV

VersionI

27

M

arc

h2

011


7/7



G

lobal


andTechnology

VolumeXIIssueIV

VersionI

28

M

arc

h2

011

This page is intentionally left blank

4 Multiple Feasible Paths in Ant Colony Algorithm

Documents

Transcript of 4 Multiple Feasible Paths in Ant Colony Algorithm