vgsim: an integrated networking and microscopic vehicular

IEEE Communications Magazine • May 20092 0163-6804/09/$25.00 © 2009 IEEE

This research is funded inpart by the National Sci-ence Foundation underthe grant number CMMI*0700383. The authorsare solely responsible forthe contents of this article.

INTRODUCTION

Vehicular ad hoc networks (VANETs) aremobile wireless networks formed by vehicleswith wireless communication and positioningcapabilities. During the last few years, VANEThas become a very popular field of research bothin academia and in industry. This is both due tothe widespread emergence of robust wirelessnetworking and positioning technologies, and,more important, the demand for the next-gener-ation intelligent transportation system to provideboth real-time traffic management and commer-cial services to vehicles on the road.

Due to the nature of mobile wireless commu-nications and the complex dynamics in real vehi-cle traffic flow, simulation is the primary tool ofchoice to analyze various applications ofVANETs. Sophisticated simulation packages areavailable for both wireless networks and vehicu-lar traffic flow; however, few of them can fullyaddress the challenging problems that arise fromthe interdisciplinary nature of VANETs. There-fore, integrating network simulation tools with

realistic vehicular traffic simulation packages isnecessary. There are different approaches tointegrating these two simulation packages. How-ever, if the underlying integrated tool cannot ful-fill all the requirements imposed by VANETapplications, the results are prone to be erro-neous or unrealistic. Therefore, a classificationof VANET applications based on simulationrequirements is necessary for accurate simula-tion design.

In general, VANET applications can be clas-sified into the following two categories:• Vehicular driver safety and traffic control

applications: These applications need toaddress the issue of how drivers respond tothe control signals disseminated using wire-less communication and the resultingchange in the topology of the underlyingVANET. Typical applications are accidentalert, real-time traffic condition update,and any applications that require drivercoordination through the VANET.

• Infotainment Applications: These applica-tions use VANET as a single- or multihopcommunication platform, and do not resultin dramatic change in the topology of theunderlying VANET. Typical applicationsinclude Internet access to vehicles, com-mercial advertisements, and various peer-to-peer applications.For both of the above classes of applications,

a network communication simulation packagewith full protocol stack support is desired. Forvehicular traffic simulation, realistic trafficmobility models are also required. For infotain-ment applications, simply integrating these twosimulation packages by using vehicular traffictraces to determine node movements in networksimulation is sufficient. However, for vehiculardriver safety and traffic control applications,real-time interactions between the network sim-ulation module and vehicular traffic simulationmodule are required.

The remainder of the article is organized asfollows. In the next section we present differentaspects of the most commonly used simulationmethodologies in VANET research (trace driv-

ABSTRACT

Simulation is the predominant tool used inresearch related to vehicular ad hoc networks. Inthis article we first present the key requirementsfor accurate simulations that arise from the vari-ous applications supported by VANETs, andreview the current state-of the-art VANET sim-ulation tools. We then present VGSim, an inte-grated networking and microscopic vehicularmobility simulation platform. VGSim providesfull-fledged wireless network simulation with anaccurate traffic mobility model. These two com-ponents are tightly integrated and can interactdynamically. We discuss the flexibility of VGSimin adopting different mobility models and alsopresent simulation results that empirically vali-date the modified mobility model we implement-ed. We discuss how VANET applications can beeasily modeled in VGSim, and demonstrate thisusing two important applications, Accident Alertand Variable Speed Limit.

AUTOMOTIVE NETWORKING SERIES

Bojin Liu, Behrooz Khorashadi, Haining Du, Dipak Ghosal, Chen-Nee Chuah, and Michael Zhang,

University of California, Davis

VGSim: An Integrated Networking andMicroscopic Vehicular MobilitySimulation Platform

LIU LAYOUT 4/10/09 5:34 PM Page 2

IEEE Communications Magazine • May 2009 3

en, open-loop, and closed-loop integration). Wethen present VGSim, an integrated VANETsimulation platform that has full-fledged net-work protocol support, a realistic microscopicvehicular traffic model, and the ability to sup-port real-time interactions between the twomodules. In the following section we first pre-sent the results that validate the mobility modelin VGSim and then discuss VGSim’s ability toadopt other mobility models. We then discusshow VANET applications are developed for sim-ulation analysis in VGSim. We then showcasevehicular driver safety applications analyzedusing VGSim. The final section concludes thisarticle.

VANET SIMULATIONMETHODOLOGIES

SIMULATION REQUIREMENTS AND DESIGNIn general, for VANET simulation there arethree dimensions in the design space: networksimulation, vehicular traffic simulation, and theintegration of these two modules. We conducteda survey of VANET research published duringthe last four years. As our survey shows, besidesin-house simulators, running a network simula-tor with traffic traces generated by a traffic sim-ulator is the main approach for VANETsimulation (more details in the next section).The traffic traces specify the vehicle mobilityduring simulation. We refer to this method asthe open-loop integration approach. The key dis-advantage of this approach is that it cannot cap-ture the dynamic interactions between theinformation exchange among the vehicles and/orroadside sensors and the traffic flow.

Another commonly used approach is theclosed-loop integration approach. In thisapproach the traffic simulator is responsible forspecifying vehicle movements throughout thesimulation process, and the network simulator isresponsible for wireless communication. Howev-er, signals transmitted using wireless channelscould be used as another type of traffic controlsignal, which can result in a change in the vehi-cles’ mobility. This is especially true for advanceddistributed multihop vehicular driver safety andtraffic control applications. In these applicationsdriver coordination based on wireless traffic con-trol signals can dramatically change drivers’behavior (accelerating, decelerating, or changinglanes), and therefore results in a traffic flow dif-ferent from what a traditional traffic simulatorcould generate. Changes in the traffic flow mayimply changes in the topology of the wireless adhoc network formed by the vehicles, which inturn can have significant impact on the perfor-mance of the wireless network. Therefore, aunique requirement for this type of VANETsimulation is the ability of capturing the “inter-actions” between wireless communication andthe vehicular mobility model. Figure 1b showsthe details of information flow and the interac-tions of the closed-loop approach.

On the other hand, not all VANET applica-tions require this “interaction” capability in theirsimulation. Infotainment applications, whichonly use a VANET as a medium to transmit

value added services such as real-time advertise-ment and Internet access service, do not neces-sarily affect the underlying topology of theVANET. If data dissemination is the only appli-cation of a VANET, the current approach ofsimple integration of a network simulator and atraffic simulator (the open-loop approach) is suf-ficient. It is, however, necessary that the adoptednetwork simulator support the entire wirelesscommunication network protocol stack to beable to carry out detailed network performanceanalysis.

In addition, since VANET simulation plat-forms are needed for evaluating potential safe-ty/infotainment applications, ease of newapplication development should also be consid-ered in the design. Simulation platforms thatadopt the approach of integrating existing trafficand network simulators may encounter complex-ities in building new VANET applications. Thisis because it requires the expertise in both simu-lation packages to build an efficient VANETapplication. Also, flexibility in adopting differentmobility models and performance issues to sup-port large-scale VANET simulations involvinghundreds or even thousands of communicationnodes (vehicles) are also important factors in thedesign of the simulation tool.

RELATED RESEARCH ONVANET SIMULATION STUDIES

We conducted a survey of VANET researchpublished during the last four years; due tospace limitation, we only highlight the most rele-vant.

As discussed above, simulation analysis of

nn

Figure 1. Information flow of two VANET simulation approaches.

NS2, Qualnet,Jist/SWANS

Vehicle traffic

Wireless communication

Network topologyVANETapps

Change

Support

Driver behavior/Mobility model

Change of velocity,position...

Change ofwirelessconnectivity

Change oflink quality,routingproperty...

Change ofwireless traffic

control signal...

Generate

Decide Decide

VISSIM, CORSIM,MANET mobility

Driver behavior/Mobility model

Support

NS2, Qualnet

Wirelesscommunication

Vehicle trafficand net

topology

(a)

(b)


IEEE Communications Magazine • May 20094

VANETs and related applications requires bothcommunication network and vehicular trafficsimulations. For the communication networksimulation, our survey shows that the majority ofresearch adopted established network simulators(NS2 [1]/QualNet [2]). For VANET simulationsthat only considered high-level communicationparameters like transmission range, some simplenetwork simulators were used. While Java inSimulation Time (JiST)/SWANS [3] is not aspopular as NS2/Qualnet, there are a number ofstudies on VANETs based on this platform [4].OMNet++ with INET Framework [5] is anoth-er platform used for simulation analysis of wire-less communication.

For vehicular traffic simulations, three typesof vehicle mobility models are typically used:• Mobility models used in mobile ad hoc net-

works (MANETs) and variants• Macroscopic vehicular traffic models• Microscopic vehicular traffic models

MANET mobility models (e.g., the RandomWay Point model) are not accurate for realisticvehicular traffic simulation, and can considerablydegrade the accuracy of the simulation results[4]. Macroscopic traffic models only specify high-level traffic metrics such as vehicle density andflow rate. A microscopic vehicular traffic model,on the other hand, specifies the behavior of eachindividual vehicle. As a result, microscopic mod-els can generally provide more realistic mobilitypatterns and detailed statistics of vehicular traf-fic flow.

For simplicity, a large body of work is basedon self-developed macroscopic traffic models.When more realistic microscopic traffic modelsare used, they are based on either high fidelitytraffic simulators such as VISSIM [6], CORSIM[7], and SUMO [8], or simulators developed bythe researchers.

Table 1 summarizes the main approachesused in VANET simulation. In the open-loopapproach with a simplistic mobility model, estab-lished network simulators like NS2 were used fornetwork simulation, and simple vehicular mobili-ty models based on a MANET or simple macro-scopic vehicular traffic models were used togenerate vehicular traffic flows [9, 10]. Open-loop means the vehicle mobility model is speci-fied at the beginning of the simulation, andunderlying attributes of vehicular traffic flowsuch as headways between vehicles and speedare predetermined and do not change as a resultof the VANET application. Figure 1a shows theinformation flow of the open-loop approach.

Specifically, the study reported in [9] describeshow to modify NS2 to accommodate this type ofVANET simulation. The open-loop approachcan also be trace-driven: vehicular traffic tracesgenerated from high fidelity commercial/non-commercial microscopic vehicular traffic simula-tors such as VISSIM [11, 16], CORSIM [12] orSUMO [15], or empirical traffic traces [13, 14]are used to describe the mobility of vehicles. Thetraffic traces specify each individual vehicle’smovement during the entire simulation. Thestudy reported in [17] is another example of thisapproach; it adopts a microscopic mobility model(IDM/MOBIL model) to determine the move-ment of the vehicles and uses OMNet++ [5] tosimulate the communication network.

Figure 1b shows the information flow for theclosed-loop approach, in which vehicular move-ments are not predetermined at the start of thesimulation. Instead, the mobility model updatesthe vehicle position, velocity, and lane in realtime based not only on the vehicular traffic flowbut also on the traffic flow control signalreceived through wireless communication. Thealtered mobility of vehicles can in turn affect thetopology of the VANET and consequently theperformance of the data communication overthe wireless network. This closed-loop informa-tion flow between the mobility model and wire-less network simulation modules cannot beprovided by the open-loop approach. Conse-quently, several studies have adopted this closed-loop approach for simulation analysis [4, 18–20].In [4] closed-loop integration is achieved by inte-grating the JiST/SWANS network simulator withStreet Random Waypoint (STRAW), a modifiedversion of the Random Waypoint mobilitymodel. This work also demostrates the impor-tance of realistic mobility models for the accura-cy of VANET simulation results. It showed thatan unrealistic mobility model of the vehicle candramatically affect the simulation results. Thestudy reported in [18] also directly supportsclosed-loop interaction between the mobilitymodel and the wireless communication module.It adopts SUMO [8] as the traffic simulator andOMNet++ [5] for wireless communication sim-ulation. The interactions between these twomodules are achieved by connecting two simula-tors through a TCP connection that is used totransfer control commands to SUMO and vehi-cle position information to the OMNet++ mod-ule. In addition, [19, 20] are also integratedVANET simulation platforms based on closed-loop integration between the two simulation

nn

Table 1. A classification of major simulation approaches in recent VANET research.

Simulation approach Description Mobility model Examples

Open-loop, simplisticmobility model

Network simulator with simplistic MANET or macroscopicmodels

MANET models or macroscopictraffic models [9, 10]

Open-loop, trace drivenmobility model

Microscopic simulation such as VISSIM generated vehicletraces fed into network simulator (e.g., NS2, QualNet) Microscopic traffic models [11–17]

Closed-loop, realisticmobility model

Integrating network communication and vehicular trafficsimulation, supporting interaction between the two Microscopic traffic models [4, 18–20]



modules. The difference between [19, 20] andother closed-loop approaches is that both net-work and vehicular traffic simulation modulesare self-developed by the authors. This, however,makes it difficult to compare with other researchbased on more well-known simulation packagessuch as NS-2 and VISSIM.

While all approaches can, to some extent, ful-fill the requirements of infotainment applica-tions, only closed-loop approaches are suitablefor accurately simulating vehicular driver safetyand traffic control applications.

VGSIM: DESIGN ANDIMPLEMENTATION

In order to easily design and analyze differentVANET applications through realistic simula-tion, we developed a highly efficient and flexibleVANET simulation platform: VGSim. VGSim isefficient in memory usage and suitable for simu-lating large-scale vehicular wireless networks. Itconsists of a network simulator with full protocolstack support, a realistic microscopic vehicularmobility model, and the closed-loop approach tointegration. VGSim’s network simulation mod-ule is based on SWANS [3], a Java-based net-work simulator. The SWANS network simulatoruses JiST, which is an event driven simulationtool [3]. The JiST simulation platform is veryefficient; it outperforms existing highly opti-mized simulation tools in both time and memoryusage. A detailed comparison of the perfor-mance efficiency of JiST/SWANS compared toother major network simulators can be found in[3]. In fact, the efficiency of JiST/SWANS makesit very suitable for VANET simulation, whichmay involve hundreds or even thousands ofsimultaneously communicating nodes.

In VGSim vehicular movements and applica-tions are transformed into events that are pro-cessed by the JiST event driven platform. Thenetwork simulator and the vehicular trafficmodel run on a feedback loop that enables theclosed-loop interaction discussed in previous sec-tions. Information obtained from the SWANSnetwork simulator is fed into the mobility modeland then based on the mobility model, updatedantenna positions are determined for theSWANS network simulator. Figure 2 shows thearchitecture of VGSim. Each entity shown inFig. 2 has a corresponding class defined in Java.Instances of the RoadEntity class represent theroad sections and hold multiple Vehicleinstances during simulation. Each Vehicleinstance mounts a radio antenna and imple-ments the wireless network communication pro-tocols defined in JiST/SWANS. Each individualobject can produce and respond to simulationevents generated by itself or other objects. Inaddition, the SWANS network simulator andvehicular mobility simulator both update agraphical interface that allows network andvehicular mobility parameters to be changeddynamically. Visualization is another importantfeature for both vehicular traffic and wirelesscommunication simulation, and many vehiculartraffic and network simulators have their ownvisualization packages [1, 6]. Enabling visualiza-

tion for vehicular traffic and wireless communi-cation at the same time in the same panel is animportant feature in VANET simulation, since itcan help in visually evaluating the correctnessand effectiveness of VANET applications. Fig-ure 3 shows a screenshot of VGSim simulating afour-lane freeway scenario with a roadside nodeand VANET enabled vehicles communicatingwith each other. It clearly shows VGSim’s visual-ization capability of overlaying communicationtraffic on top of vehicular traffic.

The vehicular mobility module of VGSim isbased on the cellular automata (CA) model,which implements a modified version of theNagel and Schreckenberg (N-S) model [21]. TheNS model is a well established CA model in traf-fic engineering research. However, in the origi-nal N-S model, the road is divided intoequal-length cells of 7.5 m, and each vehicleoccupies one cell. The simulation time granulari-ty is 1 s; hence, new vehicle positions are calcu-lated every second using the N-S model. In orderto more accurately reflect real-world traffic, wemodified the original N-S model with finer spa-tial and temporal resolution, based on the studyreported in [22]. Furthermore, we also addedlane-changing capability into our mobility model.We discuss the validation of our mobility modelin the next section.

The SWANS network simulator provides fullnetwork protocol support especially for mobilewireless communication. At the applicationlayer, SWANS provides the standard applicationnetwork interfaces. It includes both UDP andTCP protocols at the transport layer. We alsoimplemented a simple position-based routingprotocol, which leverages the GPS devices in the

nn

Figure 2. A block diagram of the VGSim architecture.

VGridApplet

RoadCanvas

RoadEntity

Vehicle

VGridApp App

SWANS network stack

Radio

Noise model RF parameters

Field

Fading Pathloss Mobility

RoadsideNode

App1 App2

Stat

isti

cs

Mob

ility



vehicles. SWANS also includes standard 802.11medium access control (MAC) layer protocoland several path loss and fading models at thephysical layer.

The introduction of close-loop interactionbetween the two simulation modules is achievedby injecting driver decision process into differentapplications. At each time step, each driver/vehi-cle makes the decision on how to change thespeed/position of the vehicle according to notonly traffic conditions perceived, but also thetraffic control messages received from the wire-less channel. Figure 4 shows the interactionsbetween the major VGSim components duringsimulation. The RoadEntity object maintainsthe main simulation loop by providing an imple-mentation of the run() method of the Proxi-able interface in JiST/SWANS, which makesthe RoadEntity a simulation entity thread inJiST/SWANS. The run() method and themoveVehicle() method of each vehicle objectare invoked at each time step. Upon invocation,each vehicle object calls the mobility modelobject’s updatePos() method to get an updatedposition information according to the mobilitymodel logic. Then the updated position informa-tion is fed into the application’s update methodupdateApp(). This method implements thelogic of the wireless communication network andtraffic control applications. At the end of eachtime step, each vehicle updates its own proper-ties such as the position for the next time step,speed limit, probability of acceleration or decel-eration, and the probability of lane changeaccording to updated application state. Theseupdated properties will result in changes in thebehavior of vehicle movement in the next timesteps. In the case shown in Fig. 4, the mobilitymodel is an implementation of the N-S model,and the Variable Speed Limit (VSL) applicationis installed on the vehicle.

MOBILITY MODEL: VALIDATION AND EXTENSION

As a vital part of VGSim, the mobility model’saccuracy determines the overall accuracy of thesimulation. In this section we first describe themodifications we made to the original N-S modeland the validation. Then we describe howVGSim can be extended to accommodate othermobility models.

VALIDATION OF THE FINER-GRAINEDN-S MODEL

In VGSim we have adopted the classic N-Smobility model used extensively in vehicular traf-fic engineering research. The original N-Smodel’s temporal-spatial resolution is adequatefor vehicular traffic engineering research. How-ever, for evaluating the performance of wirelesscommunication, the temporal resolution in termsof seconds is too coarse-grained. Therefore,updating the N-S model with a finer resolution isnecessary for accurate VANET simulation. How-ever, merely changing the resolution in the origi-nal N-S model results in inaccurate vehiculartraffic generation. Therefore, we modified theoriginal N-S model, adding more realistic accel-eration, deceleration, and lane changing behav-iors. A detailed description of the modifiedfiner-resolution N-S model is reported in [22].

In order to validate our refined fine-grainedmobility model we compared the data obtainedfrom our model with real world traffic data. Forthe latter, we used the vehicle traces producedby the NGSIM project [23]. Ideally, the moreaccurate the mobility model, the higher thedegree of correlation with the NGSIM data.

Our simulation setup consists of a five-lane700-ft (213 m) highway. In order to be able toaccurately compare with the NGSIM data, wemust guarantee that the initial and road bound-ary conditions in our simulation are the same asthose in the NGSIM data set [23]. The details ofhow we reproduce the initial and road boundary

nn

Figure 3. A snapshot of the graphical user interface of VGSim. Lines connecting vehicles show communication links between vehicles.



conditions in our simulation can be found in[22].

Our comparison is based on the fundamentaldiagram (flow-density diagram) [22]. In order toshow the accuracy of our finer resolution mobili-ty model, we first compare the fundamental dia-gram for the basic N-S models with that ofNGSIM. It is known that N-S models can pro-duce a triangular flow-density fundamental dia-gram. However, matching the fundamentaldiagram generated by the CA model with realtraffic data is a challenging task. Figure 5a showsthe fundamental diagram generated by the origi-nal N-S model (CA). It shows that in this case ofrandom slowdown noise of 0.8, the CA modelcan generate the required triangular shaped dia-gram; however, it fails to match the NGSIMdata set.

Figure 5b shows the fundamental diagramgenerated by our finer resolution model denotedfCA. This diagram shows that our finer resolu-tion model not only reproduces the classic trian-gular flow-density diagram, but also matches

with the real traffic data from NSGSIM betterthan the original N-S model. This guarantees theaccuracy of VANET simulation at higher spatialand temporal resolution.

EXTENDING VGSIM TOOTHER MOBILITY MODELS

As discussed in the previous section, commonlyadopted mobility models in vehicular trafficengineering may not completely fulfill therequirements for VANET research. Therefore,modification of common mobility models oreven incorporating a totally different mobilitymodel for better VANET simulation may berequired. Because of VGSim’s modular design,this is easily achieved by providing an implemen-tation for a mobility model interface in Java.The mobility model interface in VGSim only hasone method that must be implemented(updatePos() as shown in Fig. 4). TheupdatePos() method contains logic of how toupdate vehicle positions in any time step. There-

nn

Figure 4. Interactions among the VGSim components.

run()

r:RoadEntity

moveVehicle()

updatePos(pos:Position)

npos:Position

updateProp()

[Protocol calls]

v1:Vehicle vsl:Application Network stackNagel-Schrk:Mobility

updateApp(npos:Position)

nn

Figure 5. Comparison of the fundamental diagrams obtained using different mobility models and NGSIM data set: a) original N-Smodel (CA) vs. NGSIM; b) finer resolution N-S model (fCA) v.s. NGSIM

Density (vpm)

NGSIM vs. CA Pnoise = 0.8

0

500

Flow

(vp

h)

0

1000

1500

2000

2500

20 40 60

(a)

80 100 120

NGSIMCA

Density (vpm)

NGSIM vs. fCA Pnoise = 0.8

0

500

Flow

(vp

h)

0

1000

1500

2000

2500

50 100

(b)

150 200

NGSIMfCA



fore, extending VGSim using other mobilitymodels simply takes two steps:1. Implementing a mobility class withupdatePos() method

2. Associating each vehicle with an object ofthe mobility classAfter the simulation is executed, the

updatePos() is invoked every time step foreach vehicle. All other tasks, including placingvehicles on the road, ensuring that there are nocollisions, performing wireless communicationsimulation, and visualization, do not require anymodification in VGSim. Although VGSim is cur-rently using an in-house implementation of themobility model, it is also possible for VGSim toadopt other standalone microscopic traffic simu-lators. This is achieved by implementing aninterface wrapper for controlling and/or commu-nicating with the simulator. The ability of extend-ing VGSim to other mobility models ensuresthat VGSim is not tied to one specific mobilitymodel or vehicular traffic simulation platform.This is a limitation in many other VANET simu-lators that integrate with standalone traffic simu-lators.

VGSIM APPLICATION DEVELOPMENTAND EXAMPLES

VGSIM APPLICATION DEVELOPMENTAnother advantage of VGSim is the ease withwhich VANET applications can be developed.Due to JiST/SWANS’s flexibility of performingwireless simulation, embedding wireless simula-tion in the rest of the application logic is simplyachieved by providing a Java class with an imple-mentation of the updateApp() method (Fig. 4).Therefore, it is possible to have multiple applica-tions executing in the vehicles simultaneously. Infact, in the current VGSim, multiple applicationscan be turned on at the same time. Some appli-cations provide basic services such as positionbeaconing. Other more complex applications canmake use of the service provided by other appli-

cations such as location-aware ad hoc routing.

APPLICATION EXAMPLE: ACCIDENT ALERT ANDOPTIMAL VARIABLE SPEED LIMIT

To demonstrate the capability of our simulator,we built two VANET traffic control applications:Accident Alert (Acc Alert) and VSL on top ofVGSim. The Accident Alert application utilizesthe vehicle’s onboard wireless communicationsto send alerts to upstream vehicles of the pres-ence of an obstruction in the road ahead. Thiswill allow them to change out of impacted lanesearlier and also prevent them from changinginto those lanes. For VSL, vehicles acquire posi-tion and velocity information of other vehiclesthrough wireless communication and then coop-eratively compute the appropriate speed limitfor different sections of the road. Both applica-tions are intended to smooth vehicular traffic onhighways. Figure 6 shows the average speed vari-ance with and without the VGSim supportingAcc Alert and VSL application, with one acci-dent simulated on the road. We can see a signifi-cant decrease in variance with the use of VSLand Acc Alert. Both Acc Alert and VSL applica-tions are vehicular driver safety and traffic con-trol applications. Without VGSim’s support ofclosed-loop interaction between the networksimulation and the microscopic vehicle trafficsimulation, it is hard to evaluate the effective-ness of both applications.

CONCLUSIONSimulation is one of the most commonly usedtools in VANET studies. In this article we firstdiscuss the classification of simulation tools forVANET applications and the architecturalrequirements for accurate simulations. Afterpresenting a review of simulation tools used inVANET research, we present VGSim, which canfulfill most requirements of accurate simulation.It implements closed-loop integration of realisticvehicular traffic and a wireless communicationsimulation module. It is highly flexible and caneasily adopt different mobility models. Theapplication development process is easy and suit-able for building multiple distributed VANETapplications that can execute concurrently. Addi-tionally, since it executes as a standalone Javaapplication using the efficient JiST/SWANSpackage, it is more resource efficient thanapproaches that integrate existing network andtraffic simulators. We validate the accuracy ofthe mobility model of our simulator. Finally, wepresent results of Accident Alert and VSL asproof-of-concept applications simulated usingVGSim.

REFERENCES[1] Network Simulator 2; http://nsnam.isi.edu/nsnam/

index.php/user_information[2] QualNet; http://www.scalable-networks.com/products[3] Jist/SWANS; http://jist.ece.cornell.edu[4] D. Choffnes and F. Bustamante, “An Integrated Mobility

and Traffic Model for Vehicular Wireless Networks,”Proc. ACM VANET ’05, Sept. 2005.

[5] OMNet++; http://www.omnetpp.org[6] VISSIM; http://www.english.ptv.de/cgi-bin/index.pl[7] CORSIM; http://mctrans.ce.ufl.edu/featured/TSIS/

Version5/corsim.htm

nn

Figure 6. Average speed variance with one accident (w/ VGrid implies allvehicles implement the Acc Alert and VSL applications).

Traffic density

One accident VSL + Acc Alert variance

0.05

2

Vari

ance

(m

/s)

0

4

6

8

10

12

14

0.1 0.13 0.16 0.2 0.3 0.4 0.5 0.6 0.7

with VGridno VGrid



[8] SUMO; http://sumo.sourceforge.net/[9] A. K. Saha and D. B. Johnson, “Modeling Mobility for

Vehicular Ad Hoc Networks,” Proc. ACM VANET ’04,Aug. 2004.

[10] Y. Zhang, J. Zhao, and G. Cao, “On Scheduling Vehi-cle- Roadside Data Access,” Proc. ACM VANET ’07, July2007.

[11] M. Caliskan, D. Graupner, and M. Mauve, “Decentral-ized Discovery of Free Parking Places,” ACM VANET ’06,July 2006.

[12] J. Yin et al., “Performance Evaluation of Safety Appli-cations over DSRC Vehicular Ad Hoc Networks,” Proc.ACM VANET ’04, Aug. 2004.

[13] H.-Y. Huang et al., “Performance Evaluation of SUVnetWith Real-Time Traffic Data.,” Proc. IEEE Trans. Vehic.Tech., vol. 56, no. 6, July 2007, pp. 3381–56.

[14] D. Li et al.,” A Distance-Based Directional BroadcastProtocol for Urban Vehicular Ad Hoc Network,” Proc.Int’l. Conf. Wireless Commun., Networking, and MobileComp. 2007, Sept. 21–25, 2007, pp. 1520–23.

[15] M. Piorkowski et al., “Joint Traffic and Network Simu-lator for VANETs” MICS 2006, Zurich, Switzerland, Oct.2006.

[16] C. Lochert et al., “Multiple simulator interlinking envi-ronment for IVC,” Proc. ACM VANET ’05, Sept. 2005.

[17] C. Sommer and F. Dressler, “The DYMO Routing Proto-col in VANET Scenarios,” Proc. 66th IEEE VTC 2007-Fall,Baltimore, Maryland, Sept./Oct. 2007, pp. 16–20.

[18] C. Sommer et al., “On the Need for Bidirectional Cou-pling of Road Traffic Microsimulation and NetworkSimulation,” Proc. 1st ACM Int’l. Wksp. Mobility Mod-els Networking Research, May 2008, pp. 41–48.

[19] S. Y. Wang et al., “NCTUns 4.0: An Integrated Simula-tion Platform for Vehicular Traffic, Communication, andNetwork Researches,” Proc. 1st IEEE Int’l. Symp. Wire-less Vehic. Commun., Baltimore, MD, Oct. 2007.

[20] C. Gorgorin et al., “An Integrated Vehicular and Net-work Simulator for Vehicular Ad-Hoc Networks,” Proc.20th Euro. Simulation Modeling Conf., Oct. 2006.

[21] K. Nagel and M. Schreckenberg, “A Cellular Automa-ton Model for Freeway Traffic,” J. Physique, vol. 2,1992, pp. 2221–29.

[22] M. Zhang and H. Du, “Finer-Resolution CellularAutomata Model for Intervehicle Communication Appli-cations,” 87th Annual Meeting Transportation ResearchBoard, 2007.

[23] NGSIM; http://ngsim.fhwa.dot.gov

BIOGRAPHIESBOJIN LIU ([email protected]) is a Ph.D. student in theComputer Science Department, University of California,Davis. He received his Bachelor degree in computing fromHong Kong Polytechnic University. His research interestsinclude vehicular ad hoc networks, wireless networks, andparallel and distributed systems.

BEHROOZ KHORASHADI ([email protected]) is a Ph.D.student in the Department of Computer Science at the Uni-versity of California, Davis. He graduated from the Universi-ty of California, Berkeley in May 2004. He is currentlyworking in the Networks laboratory with a focus on peer-to-peer networks and VGrid (vehicular ad-hoc Networks).VGrid is a vehicular ad hoc networking and computing gridfor intelligent traffic monitoring and control. The goal is toevolve the intelligent transportation system (ITS) from acentralized to a distributed approach, in which vehicles cancooperatively solve traffic-flow control problemsautonomously. One example application is the lane-merg-ing scenario, especially when visibility is low during stormsor foggy weather.

HAINING DU ([email protected]) is a Ph.D. candidate in theCivil and Environmental Engineering Department at theUniversity of California, Davis. His dissertation researchattempts to develop improved microscopic vehicular trafficmodels for distributed traffic control via DSRC enabledvehicles.

DIPAK GHOSAL ([email protected]) received a B.Tech.degree in electrical engineering from the Indian Institute ofTechnology, Kanpur in 1983 and an M.S. degree in com-puter science and automation from the Indian Institute ofScience, Bangalore in 1985. He received his Ph.D. degree incomputer science from the University of Louisiana in 1988.He is currently a professor in the Department of ComputerScience at the University of California, Davis. His primary

research interests are in the areas of high-speed and wire-less networks with particular emphasis on the impact ofnew technologies on network and higher layer protocolsand applications. He is also interested in the application ofparallel architectures for protocol processing in high-speednetworks and the application of distributed computingprinciples in the design of next generation network archi-tectures and server technologies.

CHEN-NEE CHUAH ([email protected]) is an associate pro-fessor in the Electrical and Computer Engineering Depart-ment at the University of California, Davis. She received herB.S.E.E. from Rutgers University, and her M. S. and Ph.D. inelectrical engineering and computer sciences from the Uni-versity of California, Berkeley. Her research interests includeInternet measurements, network management, and wire-less/mobile computing. She is an Associate Editor forIEEE/ACM Transactions on Networking.

MICHAEL ZHANG ([email protected]) is an associate pro-fessor in the Electrical and Computer Engineering Depart-ment at the University of California, Davis. She received herB.S.E.E. from Rutgers University, and her M. S. and Ph.D. ineectrical engineering and computer sciences from the Uni-versity of California, Berkeley. Her research interests includeInternet measurements, network management, and wire-less/mobile computing. She is an Associate Editor forIEEE/ACM Transactions on Networking.


vgsim: an integrated networking and microscopic vehicular

Documents