A MODELS Noun Choi, Alieza Mahdian, Ravi Prakash ...neerajm/... · NounChoi, Alieza Mahdian, Ravi Prakash, S. Venkatesan, andNeeraj Mittal Erik Jonsson School ofEngineering and Computer

A UNIFIED FRAMEWORK OF NODE MOBILITY MODELS

Noun Choi, Alieza Mahdian, Ravi Prakash, S. Venkatesan, and Neeraj MittalErik Jonsson School of Engineering and Computer Science

University of Texas at DallasRichardson, TX

andAlbert J. Anderson, Eric Redding and Robert Butler

Performance and Architectural Collaboration Environment LabRockwell Collins, Inc.

Richardson, TX

ABSTRACT

A mobile ad hoc network (MANET) consists ofmobile nodes that communicate with each otherwithout any infrastructure. MANET has variousapplications including military operations, rescueoperations, and many other situations that requirerapidly deployable communication networks. Inorder to evaluate the performance of such net-works accurately through simulations, we needto use realistic conditions when simulating. Usermobility pattern is an important factor that af-fects the performance of the network significantly.Recently, a number of mobility models have beenproposed in order to produce realistic node mobil-ity patterns. Mobility models differ considerablydepending on the applications for which theyare designed. In this paper, we present a unifiedmobility model framework. The framework cangenerate most of widely used mobility modelswith a short script. We show this by applyingthe framework to generate most of the existingmobility models. This framework expedites theperformance evaluation process ofMANETs sinceusers do not need to write code from scratch togenerate the application-specific moving patternsfor the nodes.

INTRODUCTION

IEEE 802.11 [1] based Wireless Local AreaNetworks (WLANs) have been deployed widelyduring last few years. Combined with the emerg-ing internet technology, WLANs allow us ubiq-uitous access needed information. Wireless net-works have several significant advantages overtraditional wired networks. One type of wire-less network, namely Mobile Ad Hoc Network(MANET), is becoming popular since MANETs

can be used even when no infrastructure is avail-able and the users are mobile. While traditionaltwo-way radio, also known as walkie-talkie, pro-vides a single voice communication over a fre-quency band within an area, a MANET can carrydata communication as well as voice communi-cation with digitized audio in a secure manner.Multiple radios can communicate simultaneouslywith each other even when all the radios use thesame frequency band. Since a MANET is a packetswitched network, it also allows a user communi-cate with a designated peer or a group of userson the same frequency channel. MANETs haveapplications in military situation and search/rescueoperations. MANET can be used in many othersituations.When we develop a new device or a new

protocol for wireless networks, performance eval-uation through simulations is a critical phase. Itis difficult to develop an analytical model forMANET to estimate the performance preciselysince there are many factors that impact the perfor-mance. The correlations of those factors are oftenunknown or too complicated to model. Buildingnew devices and/or implementing new protocolsand testing them without simulating is expensivein time and budget. During simulations with var-ious parameters, we may find potential problemsand may evaluate the usefulness of the proposednew technologies. Currently, several simulators,including ns-2 [2], GloMoSim [3], and OPNET[4], are widely used to simulate MANETs.

Mobility of mobile users is an important pa-rameter that affects the performance of MANETs.User mobility patterns determine the distance be-tween communication pairs, degree of connectiv-ity of each node, the number of hidden nodes,

1-4244-1513-06/07/$25.00 ©2007 IEEE I of 7

and the network topology. These factors affect theperformance of the network significantly. Mobilitypatterns differ from one network to another basedon the end application. Simple mobility models,such as Random Waypoint Model [5], were usedin the past. More mobility models have beenproposed in order to generate a variety of mobilitypatterns. Also, more mobility models are expectedto be introduced in the future. When we test amobile network, we may need to apply variousmobility models and compare the performanceand other characteristics of the network usingthe mobility models. However, preparing mobilitypatterns for multiple mobility models is time andenergy consuming task. In this paper, we pro-pose a unified framework to generate mobilitypatterns. Our framework is a simple simulatorwhich generates the direction and the speed of thenodes at random/deterministic interval throughoutthe simulation and produces tracking log of them.A mobility model defines the rule to choose thedirection and the speed. The format of the trackinglog will be compatible with the simulator's inputmobility pattern format so that the tracking logcan be used as the input for the simulations.The framework is developed over MIT ObjectTask Command Language (OTcl) [6] and Tcl withClasses (TclCL) [7] such as the network simulatorns-2 [2]. The major components of the frameworkhave been implemented in C++ and Otcl classes.The only thing to be done by an operator, wantingto generate mobility patterns for a model, is towrite several lines of OTcl script. We presentthe structure of the framework and the way ofimplementing a mobility model in the framework.

MAJOR MOBILITY MODELS

Before we address the framework, we give abrief description of major mobility models. Forthe detailed characteristics and limitations of eachmodel, the reader is referred to [8].

Random Waypoint Mobility Model

In the Random Waypoint Mobility Model, anode chooses a destination point randomly in thesimulation field and moves toward the destinationdirectly at a speed randomly (and uniformly)chosen in the interval [mrninS, rnaxS]. On reachingthe destination, the node pauses at the position fora random amount of time, and repeats these stepsuntil the simulation finishes.

Random Walk Mobility Model

In Random Walk Mobility Model [8], a nodechooses its direction and speed randomly. Twomodes are possible with this model. In the firstmode, a node moves predefined distance before itchooses next direction and speed. In the secondmode, a node moves for predefined amount oftime.

Random Direction Mobility Model

With random waypoint model and its variants,node density is much higher at the center of thesimulation field than at the other areas [9]. Apossible somution to this problem is to let the nodemove till it reaches the border (of the simulationfield). On reaching the border, the node pausesfor a specified amount of time and continues theprocess. This is known as the Random DirectionMobility Model [9].

Gauss-Markov Mobility ModelIn the above-mentioned models, a node changes

the direction and the speed abruptly. In the realworld, this is unlikely to happen. The Gauss-Markov Mobility Model [10] assumes that thevelocity of a node is correlated over time thusyielding a model where a node changes the di-rection and the speed in a smooth way. In a 2-dimensional plain, direction and the speed of anode at time t can be represented with a velocityvector Vt = [v< i]T, which is recalculated at agiven interval. The velocity at t can be computedby:

x= av1 4+ (1 -))v +--(1- a2)W_

VY = avy 1 + (1 -a)vY + ( a2) 1(1)

where a (in the range [0, 1]) is the smoothingfactor, vx and vY are asymptotic means of thevelocity, and wx1 and wty_ are Gaussian randomvariables with zero mean and variance of ax and&Y respectively.

City Section Mobility ModelIn the City Section Mobility Model [11], nodes

can move only along the roads and the streets ina city.

Reference Point Group Mobility ModelIn all of the models reviewed so far, a node

independently selects direction and speed (inde-pendent of the whereabouts of the other nodes).

2 of 7

Certain types of mobile users, such as army sol-diers in a squad or members of a rescue operation,may form a group and move together. Severalgroup mobility models have been proposed in or-der to depict various situations. Many other groupmobility models can be implemented as a vari-ant of Reference Point Group Mobility (RPGM)Model [12]. In RPGM, each group has a logicalcenter. The logical center of a group is used tocalculate the movement of the group. Each nodein the group has a reference point whose relativelocation to the group logical center is predefined.RP(t) represents the location of the referencepoint at time t. RP(t + 1) can be calculated basedon moving vector, GM, of group logical center.Once RP(t + 1) is calculated, a node combinesRP(t + 1) with a random motion vector, RM, ofthe node to calculate the target of the node at timet + 1.

0

0

ode

RP(t) /+

"' 0

Fig. 1. Reference Point Group Mobility Model

provides links between OTcl objects and C++objects. The framework is a simple event-drivensimulator as well as an OTcl interpreter. Inputscript to the framework consists of several parts:First part of the script provides a list of requiredparameters. The script has a part to create neces-sary objects, which includes a Framework object,a Scheduler object and a number of MobileNodeobjects. After creating objects, the script initializesthe objects and initiates the simulation. Finally,the direction and speed selection rule of a node inaccordance with the mobility model is to appearin the script as a member method of MobileNodeclass. The name of the method is select-target.Once the simulation is initiated, nodes producetracking log whenever they change the directionand/or speed. The format of the tracking log iscompatible with the designated simulator. At themoment, the framework generates ns-2 compatibletracking log and we plan to extend this to severalmore simulators including OPNET [4] and Glo-MoSim [3].

Scheduler

1) F ItchEvent\

2) Dispat the Event\ ~~~~~~Sorted Event Queue

5) Schedule Nex Mo}vA

Mobil d MobileNode 2 bMoileNode-k Mobid d n

4 //dat oca

select-targcneslet-trgtt

Fig. 2. Basic Structure of Mobility Framework

UNIFIED FRAMEWORK

We design the framework with three objec-tives: 1) Generating mobility patterns for a modelshould require as little work as possible. 2) Theframework should be able to generate mobilitypatterns for most, if not all, well-known mobilitymodels. 3) It should be broad enough that a newly-introduced model can be easily implemented onthe framework. With these objectives, we decidedto make the framework script-powered so that anoperator can easily implement new mobility modelon the framework by writing a few lines of scripts.We developed the framework using OTcl [6] andTclCL [7]. OTcl is an Object-Oriented extensionof Tool Command Language (Tcl) [13] and TclCL

Fig. 2 depicts the simplified structure and work-flow of the framework. On creation, a MobileNodeobject reserves with the Scheduler the first mobileparameter selection event at time t before the sim-ulation begins. The events are sorted in time at theevent queue. After being initiated, the Schedulerfetches the Head of Line (HOL) event from thesorted event queue and dispatches the event to thecorresponding object, one of MobileNode objects.On event's arrival, the MobileNode calculates itscurrent location using the information of last up-date time, direction, speed, and the coordinatesof the destination. It then calls member methodcalled select-target, which is provided as the inputto the framework as a part of the script. select-

3 of 7

.

target defines the rule to change direction and/orspeed of the node in accordance with the model.After selecting the parameters (such as directionor destination point, speed, and duration of themovement), it adds 'the next parameter selectionevent' to the scheduler. The scheduler stores theevent in the sorted event queue. When a nodechooses the next direction (or target location) andthe speed, it produces tracking log unless the nodeis a virtual node. A virtual node works in a manner

identical to an ordinary node except that it doesnot produce tracking log.

The important member variables and theirmeanings for the MobileNode class are listed inTABLE I. Note that other member variables can

be easily added to a class with OTcl when theyare needed. We do not need to declare membersof a class in advance with OTcl. Instead, creatingan instance of a variable in any method using setor instvar in a method body makes the variable a

member of the class [6]. This feature allows us toextend the MobileNode class easily as required.

TABLE I

VARIABLES OF MOBILENODE

The framework provides a range of librarymethods. A few random variable generating func-tions are available including uniform, exponential,and normal random variables. Interfaces to access

the scheduler from the script are available. Thus,events can be added or deleted through the script,if necessary. Important Member methods of Mo-bileNode class are explained in the following.

target-random-point: returns coordinates ofrandomly selected point in the simulationplaintarget-random-vector: returns coordinates ofthe point which is the result of vector transi-tion from current position. Random directionand given distance decide the vectortime-to-dest: returns required time to the des-tination point given the speedspeed-to-reach-dest: returns the speed re-

quired to reach the destination in a given timeapprox-point: returns true when given twonumbers differ less than predefined value.Used when a node reaches the border of thesimulation fieldget-dist: calculates the distance from the cur-

rent location to a given pointreference-node: sets the reference node. Usedfor group mobility modelsguess-position: returns predicted location ofthe node at a given time instanceupdate-position: calculates current location ofthe noderandom-pos: place the node at a randomlyselected locationset-position: place the node at a given pointsetdest: set destination coordinates and speedand schedule the next destination selection

IMPLEMENTATION OF MODELS

This section presents how to implement mobil-ity models on the framework and shows examplesof mobility patterns that were produced by theframework. As we explained in the previous sec-

tion, writing model-dependent MobileNode class'method, select-target, is what we need to do inmost cases.

Random Waypoint Mobility Model

Initially deploy nodes randomly. On select-target, choose a random point as the next des-tination. Use a uniform random variable s in[minnS, maxS] for speed and uniform random

4 of 7

Name Usage

index Id of the node

X X coordinate at the most recent

position update

Y Y coordinate at the most recent

position update

destX Target X coordinate

destY Target Y coordinate

vX Velocity for X axis

vY Velocity for Y axis

speed Current speed

lastUpdate Time at which the position was updated

schedule Time reserved for the next target selection

predictX Expected coordinate at time t

Used with method guess-position

predictY Expected coordinate at time t

Used with method guess-position

variable p in [mInjP, maxP] for pause time, whereminS, maxS, minP, and maxP are given asparameters. Finally calculate next destination se-lection time with tc,ur + ttd + p, where tc,ur isthe current time and ttd is the amount of timerequired to reach the destination. Fig. 3 showsthe mobility pattern of a node produced by theRandom Waypoint Model.

o0 500 1000 1500 2000

200

400

600

o0 500 1000 1500 2000

200

400

600

800\

1000

1200

1400

1600

1800

2000

Fig. 4. Random Walk with Constant Distance800

1000

1200-

1400-

1600-

1800-

2000

Fig. 3. Mobility pattern of Random Waypoint

Random Walk Mobility Model

Two modes are available for Random WalkMobility Model: 1) A node can move constant dis-tance before it changes the speed and the direction.2) A node can move for constant amount of timebefore it chooses another destination. Nodes arerandomly placed initially. On select-target, choosea random speed s from uniform distribution. Ifthe mode is constant time mode, the distance iscalculated with s x T, where T is given time inter-val. In the constant distance mode, the distance isgiven. The movement vector can be achieved bychoosing random direction (angle in radian) fromuniform distribution. Next destination selectiontime is calculated as ttd. When a node movestoward a point outside of the simulation area, thenode bounces off at the border and travels theremaining distance. Produced mobility pattern forconstant distance mode is shown in Fig. 4.

Random Direction Mobility Model

Initially, nodes are placed randomly. Whencalling the method select-target, a node choosesrandom speed and direction using uniform dis-tribution. If the node is at the border and cho-sen direction is toward outside of the simulation

area, turn the direction 180°. The node trav-els until it reaches another border. The way toget the distance to the destination is as fol-lows: Calculate distances to all four borderswith (minX- X)/cos(0), (maX-X)/cos(0),(minY - Y)/siln(O), and (maxY - Y)l/sin(O),where minX, maxX, minY, and maxY arecoordinates of borders, and X and Y are co-ordinates of the current location, and 0 is thechosen angle. Drop negative values in computingdistances. Indeed two of them should be no largerthan zero since those borders are at the oppositedirection from the chosen angle. Finally choosethe minimum values from the remaining ones.Next target scheduling can be done in a similarway as Random Walk Model. Fig. 5 shows theproduced mobility pattern.

o0 500 1000 1500 2000

200"I

400

600

800-

1000

1200-

1400-

1600-

1800-

2000 \

Fig. 5. Mobility Pattern for Random Direction

5 of 7

Gauss-Markov Mobility ModelEquation (1) gives the rule. Speed and destX

and destY is calculated as followings:

St = ut x Yt+ 1j x V

destXt =Xt1 + v x T

destYt = Y1-I + vy x T

(2)

o0 500 1000 1500 2000

200

400

600

800

1000

1200

where Xt 1 and Yt I are coordinates of currentlocation and T is the time interval for the targetselection.

With Gauss-Markov Model, asymptotic meansv' and vY have significant impact on the direc-tion. Thus, whenever a node reaches the borderwe change those values with randomly selectedvariables from normal distribution. Fig. 6 showsthe produced mobility pattern.

o0

200

400

600

800

1000

1200

1400 9

1600

1800

2000

Fig. 6. N/

City Section M

In order to

1400

1600

1800

2000

Fig. 7. Mobility Pattern for City Section Model

Reference Point Group Mobility Model

A group in Reference Point Group Model500 1000 1500 2000 (RPGM) has a virtual node to represent the log-

ical center of the group. The logical center is/ 0 initially randomly deployed. Each group member

/ randomly chooses offset vector for its own refer-/ t ence point and places itself at the reference point.

On select-target at time t, virtual nodes followtarget selection process for Random Walk Modelwith constant interval. A group member uses thepredicted location of the logical center, i.e. thevirtual node for the group, at time t+ 1 and locatesits own reference point, RP(t+ 1), accordingly. Asmall random motion vector, RM, with RP(t+ 1)at the origin of RM decides the destination point

lobility Pattern for Gauss-Markov of the node. We need to make sure that logicalcenter makes target selection before any member

obility Model of the group. Fig. 8 shows the resulting mobilityimplement City Section Mobility pattern when there are five members in a group.

Model easily, we add a class called Intersect. AnIntersect object represents an intersection of roadand street. Each Intersection object has a list ofintersections that have direct path to the intersec-tion; In other words, each object in the list canbe reached from the object under considerationwithout changing direction. Each node is initiallylocated at a randomly selected intersection. Onselect-target the node chooses a destination fromthe list of reachable intersections and choosesrandom speed. Next destination selection time isdecided with the speed and the destination. Fig.7 shows the produced mobility pattern of singlenode where there are 16 intersections which arelocated at (400i, 400j) and i and j go 1 through4.

o0

200

400

600

800

1000

1200

1400 $1600

1800

2000

500 1000 1500 2000

Fig. 8. Mobility Pattern for RPGM

6 of 7

CONCLUSION

In this paper, we have introduced a unifiedframework for mobility models for mobile ad hocnetworks. In order to make the framework flexibleand applicable to possible future models that maybe developed, we develop the framework overMIT OTcl and TclCL. The framework is script-powered. A MobileNode object changes destina-tion and speed at random or deterministic intervalsuntil the simulation terminates. Whenever a nodechanges direction and/or speed, the node producesa tracking log whose format is compatible withthe designated simulator. Model-dependent desti-nation and speed selection rule is to be definedin a member method of MobileNode class calledselect-target. We have demonstrated that severalmobility models can be easily implemented us-ing the framework and the framework producesmodel conforming mobility patterns. Since theframework provides skeleton of mobility patternproducing process and a range of library meth-ods, implementing new mobility models using theframework will require considerably less amountof time and effort than writing codes for themodels from scratch.

10. B. Liang and Z. J. Hass, "Predictive distance-basedmobility management for PCS networks," in Proc. ofIEEE Information Communications Conference (INFO-COM 1999), New York, NY, Apr. 1999.

11. J. Tian, J. Hahner, C. Becker, I. Stepanov, and K. Rother-mel, "Graph-based mobility model for mobile ad hocnetwork simulation," in Proc. of35th Annual SimulationSymposium, San Diego, CA, Apr. 2002.

12. X. Hong, M. Gerla, G. Pei, and C. Chiang, "A groupmobility model for ad hoc wireless networks," in Proc.of the ACM International Workshop on Modeling andSimulation of Wireless and Mobile Systems (MSWiM),Seattle, WA, Aug. 1999.

13. J. K. Ousterhout, Tcl and the Tk Toolkit. Addison-Wesley, 1994.

REFERENCES

1. IEEE 802.11 WG, "IEEE standard for information tech-nology - LAN/MAN - specific requirements - part 11:Wireless LAN medium access control (MAC) and phys-ical layer (PHY) specifications," IEEE Std 802.11-1999,1999.

2. K. Fall and K. Varadhan. (2007) The ns manual.[Online]. Available: http://www.isi.edu/nsnam/ns/doc/

3. M. Takai, L. Bajaj, R. Ahuja, R. Bagrodia, and M. Gerla,"Glomosim: A scalable network simulation evironment,"Technical Report 990027, 1999.

4. (2007) OPNET technologies.,inc. [Online]. Available:http://www.opnet.com

5. D. Johnson and D. Maltz, Dynamic source routing in adhoc wireless networks, T. Imelinsky and H. Korth, Eds.Kluwer Academic Publishers, 1996.

6. D. Wetherall and C. Lindblad, "Extending tcl for dy-namic object-oriented programming," in Proc. of theTcl/Tk Workshop '95, Toronto, CANADA, July 1995.

7. (2007) Tclcl. [Online]. Available: http:Hotcl-tclcl.sourceforge.net/tclcl/

8. T. Camp, J. Boleng, and V. Davies, "A survey of mobilitymodels for ad hoc network research," Wireless Commu-nications and Mobile Computing, vol. 2, no. 5, pp. 483-502, 2002.

9. E. M. Royer, P. M. Melliar-Smith, and L. E. Moser, "Ananalysis of the optimum node density for ad hoc mobilenetworks," in Proc. ofthe IEEE International Conferenceon Communications (ICC), Helsinki, Finland, June 2001.

7 of 7