On the Need for Bidirectional Coupling of Road Traffic Microsimulation and Network Simulation Christoph Sommer, Zheng Yao, Reinhard German and Falko Dressler Computer Networks and Communication Systems Department of Computer Science, University of Erlangen, Germany {christoph.sommer,german,dressler}@informatik.uni-erlangen.de ABSTRACT Simulation of network protocol behavior in Vehicular Ad Hoc Network (VANET) scenarios is strongly demanded for evaluating the applicability of developed network protocols. In this work, we discuss the need for bidirectional coupling of network simulation and road traffic microsimulation for eval- uating such protocols. The implemented mobility model, which defines all movement of nodes, influences the out- come of simulations to a great deal. Therefore, the use of a representative mobility model is essential for producing meaningful results. Based on these observations, we devel- oped the hybrid simulation framework Veins (Vehicles in Network Simulation), composed of the network simulator OMNeT++ and the road traffic simulator SUMO. Based on a proof-of-concept study, we demonstrate the advantages and the need for bidirectionally coupled simulation. Categories and Subject Descriptors C.4 [Performance of Systems]: Modeling Techniques; C.2.1 [Computer-Communication Networks]: Network Architecture and Design—Wireless Communication ; I.6.m [Simulation and Modeling]: Miscellaneous General Terms Measurement, Performance Keywords Vehicular ad hoc networks, network simulation, road traffic microsimulation 1. INTRODUCTION In this paper, we investigate the need for bidirectional coupling of network and road traffic simulation for more realistic Vehicular Ad Hoc Network (VANET) simulation experiments. The development of adequate Inter Vehicle Communication (IVC) protocols using VANETs is in the Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. MobilityModels’08, May 26, 2008, Hong Kong SAR, China. Copyright 2008 ACM 978-1-60558-111-8/08/05 ...$5.00. Figure 1: Overview of the coupled simulation frame- work. State machines of road traffic and network simulator communication modules. main focus of such simulations [2], e.g. for incident detection such as traffic jam and accident detection. In the following, we motivate the demand for more sophisticated simulation techniques by investigating the state of the art in network and road traffic simulation. We developed a special simulation framework that pro- vides coupled network and road traffic simulation using well- established simulators from both communities. In particu- lar, we employ OMNeT++ 3.4b2 [23], a simulation envi- ronment free for academic use, for modeling realistic com- munication patterns of VANET nodes. Traffic simulation is performed by the microscopic road traffic simulation package SUMO [10]. Developed by German research organizations DLR and ZAIK, this simulator is in widespread use in the research community, which makes it easy to compare results from different network simulations. Availability of both sim- ulators’ C++ source code under the terms of a GPL license made it possible to integrate all needed extensions into the respective simulation cores. An overview of the resulting, coupled simulation framework, which we named Veins 1 (Ve- hicles in Network Simulation), is given in Figure 1. Further- more, we study the applicability of bidirectionally coupled network and road traffic simulation using a sample scenario evaluating the influence of IVC on road traffic. 1 http://www7.informatik.uni-erlangen.de/veins/
Transcript
On the Need for Bidirectional Coupling ofRoad Traffic Microsimulation and Network Simulation
Christoph Sommer, Zheng Yao, Reinhard German and Falko DresslerComputer Networks and Communication Systems
Department of Computer Science, University of Erlangen, Germany{christoph.sommer,german,dressler}@informatik.uni-erlangen.de
ABSTRACTSimulation of network protocol behavior in Vehicular AdHoc Network (VANET) scenarios is strongly demanded forevaluating the applicability of developed network protocols.In this work, we discuss the need for bidirectional coupling ofnetwork simulation and road traffic microsimulation for eval-uating such protocols. The implemented mobility model,which defines all movement of nodes, influences the out-come of simulations to a great deal. Therefore, the use ofa representative mobility model is essential for producingmeaningful results. Based on these observations, we devel-oped the hybrid simulation framework Veins (Vehicles inNetwork Simulation), composed of the network simulatorOMNeT++ and the road traffic simulator SUMO. Basedon a proof-of-concept study, we demonstrate the advantagesand the need for bidirectionally coupled simulation.
Categories and Subject DescriptorsC.4 [Performance of Systems]: Modeling Techniques;C.2.1 [Computer-Communication Networks]: NetworkArchitecture and Design—Wireless Communication; I.6.m[Simulation and Modeling]: Miscellaneous
General TermsMeasurement, Performance
KeywordsVehicular ad hoc networks, network simulation, road trafficmicrosimulation
1. INTRODUCTIONIn this paper, we investigate the need for bidirectional
coupling of network and road traffic simulation for morerealistic Vehicular Ad Hoc Network (VANET) simulationexperiments. The development of adequate Inter VehicleCommunication (IVC) protocols using VANETs is in the
Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.MobilityModels’08, May 26, 2008, Hong Kong SAR, China.Copyright 2008 ACM 978-1-60558-111-8/08/05 ...$5.00.
Figure 1: Overview of the coupled simulation frame-work. State machines of road traffic and networksimulator communication modules.
main focus of such simulations [2], e.g. for incident detectionsuch as traffic jam and accident detection. In the following,we motivate the demand for more sophisticated simulationtechniques by investigating the state of the art in networkand road traffic simulation.
We developed a special simulation framework that pro-vides coupled network and road traffic simulation using well-established simulators from both communities. In particu-lar, we employ OMNeT++ 3.4b2 [23], a simulation envi-ronment free for academic use, for modeling realistic com-munication patterns of VANET nodes. Traffic simulation isperformed by the microscopic road traffic simulation packageSUMO [10]. Developed by German research organizationsDLR and ZAIK, this simulator is in widespread use in theresearch community, which makes it easy to compare resultsfrom different network simulations. Availability of both sim-ulators’ C++ source code under the terms of a GPL licensemade it possible to integrate all needed extensions into therespective simulation cores. An overview of the resulting,coupled simulation framework, which we named Veins1 (Ve-hicles in Network Simulation), is given in Figure 1. Further-more, we study the applicability of bidirectionally couplednetwork and road traffic simulation using a sample scenarioevaluating the influence of IVC on road traffic.
The contributions of this paper can be summarized as fol-lows. We present such a means of bidirectional coupling,which allows the network simulation to directly control theroad traffic simulation and thus to simulate the influenceof VANET communications on road traffic. Based on theSUMO road traffic microsimulation tool and the OMNeT++network simulation framework, we developed the integratedVANET simulator Veins that allows dynamic interaction be-tween both simulators (Section 5). As a proof of concept forincident warnings, we used the coupled simulation frame-work to evaluate two mechanisms for incident warnings andtraffic jam prevention (Section 6).
2. NETWORK SIMULATIONNetwork simulation is commonly used to model computer
network configurations long before they are deployed in thereal world. Through simulation, the performance of differentnetwork setups can be compared, making it possible to rec-ognize and resolve performance problems without the needto conduct potentially expensive field tests. Network simu-lation is also widely used in research, in order to evaluatethe behavior of newly developed network protocols [6].
In most cases, network protocols are analyzed using dis-crete event simulation and a large number of simulationframeworks is available in this domain. Examples of suchframeworks are open source tools such as the network sim-ulator ns-2 [3], OMNeT++ [23], J-SIM [16], and JiST /SWANS [1] and commercial tools like OPNET. The work-ing principles of all these simulators are similar and the dif-ferences lie mostly in the number of available models, i.e.typical MAC and routing protocols. Also, the support forlarge node numbers varies.
In our evaluation, we are using the network simulatorOMNeT++, together with its INET Framework extension,for simulating VANET protocols. Thus, in this section, weprovide a short overview to event-based network simulationusing OMNeT++. Without losing generality, it can be saidthat similar techniques are used by the other simulationtools as well.
Scenarios in OMNeT++ are represented by a hierarchyof reusable modules written in C++. Modules’ relation-ships and their communication links are stored as NetworkDescription (NED) files and can be modeled graphically.Simulations are either run interactively, in a graphical en-vironment, or are executed as command-line applications.The INET Framework provides a set of OMNeT++ modulesthat represent various layers of the Internet protocol suite,e.g. the TCP, UDP, IPv4, and ARP protocols. It also pro-vides modules that allow the modeling of spatial relations ofmobile nodes and IEEE 802.11 transmissions between them.
Mobility support for network simulations is usually lim-ited to simple mobility patterns. Examples that are availablein almost all network simulation frameworks are the Ran-dom Waypoint or Manhattan mobility models. It is widelyaccepted that such simple mobility patterns cannot be usedfor experiments in VANET scenarios as road traffic patternsstrongly differ from such simple mobility models.
3. TRAFFIC MICROSIMULATIONStrictly speaking, for the most realistic simulation of mov-
ing nodes, their mobility would need to be deduced fromtrace files obtained in real-world measurements. However,
even if such trace files could be readily created for a specificscenario, simulations could still only be performed for ex-actly the scenario one was able to gather movement tracesfor. Varying only a single parameter, e.g. traffic density, andkeeping all other parameters unchanged, would be infeasi-ble with this approach. Full control over all aspects of thescenario can, however, be achieved if movement traces aregenerated by traffic simulation tools.
Traditionally, road traffic simulation models are classifiedinto Macroscopic, Mesoscopic, and Microscopic models, ac-cording to the granularity with which traffic flows are exam-ined. Macroscopic models, like METACOR [5], model trafficat a large scale, treating traffic like a liquid and often ap-plying hydrodynamic flow theory to vehicle behavior. Meso-scopic models, like CONTRAM [20], are concerned with themovement of whole platoons, using e.g. aggregated speed-density functions to model their behavior. Simulations ofVANET scenarios, however, are concerned with the accu-rate modeling of single radio wave transmissions betweennodes and, therefore, require exact positions of simulatednodes. Both Macroscopic and Mesoscopic models cannotoffer this level of detail, so only Microscopic simulations,which model the behavior of single vehicles and interactionsbetween them, will be considered as mobility models for sim-ulated VANET nodes.
Transportation and traffic science has developed a numberof microsimulation models, each taking a different approachand thus each resulting in simulations of different complex-ity. Models that are in widespread use within the trafficscience community include the Cellular Automaton (CA)model [14], the SK car-following model developed by Ste-fan Krauß [11], as well as the IDM/MOBIL model [21, 22].When doing traffic simulation, each approach has its partic-ular advantages and particular drawbacks. However, the ac-curacy of many of these models was evaluated in [4], whichconcluded with the recommendation to “take the simplestmodel for a particular application, because complex mod-els likely will not produce better results”. Essentially thismeans that, as far as network simulation is concerned, allcommon microsimulation approaches are of equal value as amobility model.
Today, several simulation environments exist which cangenerate trace files of vehicles moving according to thesemicrosimulation models. Common tools include Daimler-Chrysler’s FARSI or VISSIM by PTV AG. In the interestof comparability of research results, however, it is evidentlymore beneficial to use readily available simulation environ-ments, as using the same mobility model is the easiest andsometimes the only way of accurately reproducing resultsobtained in related work.
Traffic simulation in Veins is performed by the micro-scopic road traffic simulation package SUMO, which usesthe aforementioned SK mobility model, can perform simu-lations both running with and without a GUI and importscity maps from a variety of file formats. SUMO allows high-performance simulations of huge networks with roads con-sisting of multiple lanes, as well as of intra-junction traffic onthese roads, either using simple right-of-way rules or trafficlights. Vehicle types are freely configurable with each vehiclefollowing statically assigned routes, dynamically generatedroutes, or driving according to a configured timetable.
The use of such microscopic road traffic simulation in com-bination with IVC protocol analysis using a state-of-the-art
Message Exchange
SUMOOMNeT++/INET
1: SIM_NODE_STOP
2: SIM_NODE_REROUTE
3: SIM_NODE_RESUME
4: SIM_STEP
Se
nd
Co
mm
an
ds
Trig
ger
SU
MO
Tim
est
ep
Ad
van
ce
Sim
ula
tio
n
5: Trace Data
Bu
ffe
r C
om
ma
nd
s
Figure 2: Messages exchanged between road trafficand network simulator communication modules.
network simulator can provide deeper insights into the be-havior of VANET protocols than is possible with one alone.This is especially the case if IVC can directly influence theroad traffic, e.g. through incident warnings or other trafficmessages. Such an evaluation requires a bidirectional cou-pling of both simulators.
4. RELATED WORKTraditionally, the mobility models used in many network
simulation tools do not take into account driver behavior orspecific characteristics of the urban environment (presenceof stop lights, intersections, merge lanes, etc). As a result,the simulation of network protocols may be unrealistic.
One major advancement in this domain was the conceptof trace-based mobility modeling to be used in network sim-ulation environments. Here, realistic mobility patterns aregenerated (off-line) and used as representative models forthe evaluation of network protocols. In fact, as a commonpractice in many simulation platforms, the mobility tracesare normally inserted into network simulation modules asindependently-generated off-line files. This way, the systemcomplexity is reduced. Two methods for the generation oftrace files can be distinguished. First, real-world observa-tions can be used, i.e. the mobility of real vehicles is observedin a city or highway environment and the resulting trace in-formation is processed for use in network simulations [7,12].Similarly, mobility patterns can be extracted from these real-world observations to analytically model traffic flows [25].
Another approach is to employ traffic microsimulationtools coupled with network simulators. An early exampleis based on the integration of VISSIM traces with the net-work simulator ns-2 [13], a frequently used simulation frame-work. Similarly, the SUMO traffic microsimulation tool hasbeen integrated with ns-2, resulting in the hybrid simulationframework TrANS [15].
Hybrid simulation and mathematical modeling have re-cently been combined in order to speed up the simulationprocess [9]. Also, our preliminary work facilitated coupledsimulation using a road traffic microsimulation model basedon the IDM/MOBIL model together with the OMNeT++network simulation tool [18].
Figure 3: Excerpt of the movement trace, as sent bythe road traffic simulator.
Nevertheless, such “de-coupling” design philosophy facesone dilemma: If the results from the network simulationmodule can affect the mobility trace, this off-line “isolated”methodology is unable to generate the real-time interactionbetween the mobility model simulation module and the net-work simulation module. For example, in vehicular safetyapplications, vehicles will generate alert messages to changethe mobility patterns of other vehicles. In this case, the net-work simulation model and the mobility simulation modelneed to interact with each other in a real-time manner.
This problem has been addressed with the NCTUns sim-ulation environment [24]. This tool is similar to TrANS butallows integrated network and traffic simulation. The mainproblem of this tool, which has been developed from scratch,is that the models in both domains (network and road trafficmicrosimulation) are hard to compare to well-tested mod-els using standard simulation environments. Additionally,the manifold implementations of models for various networkprotocols, available for e.g. ns-2 or OMNeT++, cannot beused.
5. BIDIRECTIONAL COUPLINGWe achieved bidirectional coupling of both frameworks,
the network simulator OMNeT++ and the road traffic sim-ulator SUMO, by extending each with a dedicated commu-nication module. During simulation runs, these communi-cation modules exchanged commands, as well as mobilitytraces, via TCP connections.
OMNeT++ is an event-based simulator, so it handles mo-bility by scheduling node movements at regular intervals.This fits well with the approach of SUMO, which also ad-vances simulation time in discrete steps. As can be seen inFigure 1, the control modules integrated with OMNeT++and SUMO were able to buffer any commands arriving in-between timesteps to guarantee synchronous execution atdefined intervals. At each timestep, OMNeT++ would thensend all buffered commands to SUMO and trigger the cor-responding timestep of the road traffic simulation. Uponcompletion of the road traffic simulation timestep, SUMOwould send a series of commands and the position of all in-stantiated vehicles back to the OMNeT++ module. Afterprocessing all received commands and moving all nodes ac-cording to the mobility information, OMNeT++ would thenadvance the simulation until the next scheduled timestep,allowing nodes to react to altered environment conditions.
(a) OMNeT++/INET (b) SUMO
Figure 4: Screenshots of simulators’ graphical user interfaces running network and road traffic simulations inparallel.
Figure 2 shows the commands sent by OMNeT++ tothe road traffic simulator, allowing it to influence vehicles’behavior. Using these commands, vehicles in SUMO canbe stopped, resumed, and rerouted to avoid arbitrary roadsegments. Also illustrated in Figure 2 are the alternatingtwo phases of coupled simulation which result from this ap-proach. In the first phase, commands are sent to SUMO, inthe second phase their execution is triggered and the result-ing mobility trace received.
Figure 3 shows a small sample of the command and mo-bility trace stream sent by SUMO to the network simula-tion. To guarantee synchronicity of both simulators, eachtimestep is signaled by one tsp command containing thecurrent simulation time. Using the add command, the roadtraffic simulation is able to introduce new vehicles enteringthe road traffic simulation, to be represented by an arbitraryOMNeT++ module. In the example shown, new modulesof type “Car” using images “car0 vs” and “car1 vs” are in-stantiated at timesteps 0 and 8. Similarly the road trafficsimulation is able to remove from the network simulation allvehicles that have reached their destination by issuing del
commands. Mobility traces are communicated by transmit-ting the current speed and position of all instantiated nodesas a series of mov commands, the position being expressedas both OMNeT++ simulation coordinates and SUMO laneidentifier.
Figure 4 shows screenshots of the GUI versions of bothsimulators running a coupled simulation of IVC in trafficstreams merging at an intersection.
6. PROOF-OF-CONCEPT EXAMPLESIn the following, we demonstrate the advantages of bidi-
rectionally coupled network and road traffic simulation usingVeins in a proof-of-concept example. In particular, we usethe scenarios depicted in Figure 5 for evaluating the influ-ence of IVC protocols on road traffic using bidirectionallycoupled simulators. Detailed information on the used pro-tocol and an in-depth performance evaluation can be foundin [19].
6.1 First Scenario and SetupThe scenario used in our simulations consists of a number
of simulated single-lane roads. The roads are laid out in agrid with a cell size of 1 km. Simulations are performed forgrid sizes ranging from 5×5 roads to 16×16 roads. In eachsimulation, all vehicles start, one by one, at a fixed sourcenode in the top left corner of the grid. If no IVC takes placevehicles then travel along the shortest route to a fixed sinknode located in the bottom right corner of the grid.
Traffic obstructions are introduced by stopping the leadvehicle for 60 s or 240 s, depending on the scenario. As eachroad offers a single lane per driving direction, nodes cannotovertake each other and, hence, need to find a way aroundblocked roads by means of IVC, or get stuck in traffic.
To provide ad hoc routing among the nodes, we use ourimplementation [17] of the Dynamic MANET On Demand(DYMO) routing protocol as an application-layer module ofthe INET Framework module set. As per the specification,it uses a node’s UDP module to communicate with otherinstances of DYMO, to discover and maintain routes andthus establish a VANET.
(a) UDP IVC scenario. Com-munication relies on VANETalone.
(b) 1000 vehicles on a 16×16 grid; lead vehicle stops for 240 s
Figure 6: Average speed of individual vehicles, ordered by time of departure. One scenario with free flowingtraffic, one scenario with incident and IVC. Vehicles poll the Traffic Information Center every minute.
Two different types of IVC, illustrated in Figure 5, areexamined. In both scenarios vehicles with a speed of zero,after some time, start to inform other vehicles of a potentialincident on the current lane, causing them to avoid this lane.When the originating vehicle resumes its journey, it notifiesother vehicles that the lane can be used again.
Figure 5(a) displays the UDP scenario, in which this noti-fication was realized by flooding incident warnings throughthe VANET over 5 hops or 25 hops, depending on the sce-nario. Upon receiving an incident warning, a vehicle wouldquery the originating node if the warning was current and,if it received a positive reply, try and avoid the lane in ques-tion.
Figure 5(b) displays the TCP scenario, in which a numberof Roadside Units (RSUs), each connected to a central traf-fic information service, were added to each intersection tosupport IVC. In this scenario, vehicles maintained a TCPconnection to the central server, which was used to pub-lish and revoke incident information. In intervals of 60 s or180 s, depending on the scenario, vehicles also used the TCPconnection to retrieve a list of incident warnings from thecentral server.
Table 1 lists the values used to parameterize the vehiclesof the road traffic microsimulation, modeling dense inner-city traffic with inattentive drivers. We configured vehiclesto drive at a maximum speed of 14 m/s and modeled denseinner-city traffic with inattentive drivers.
Table 1: Road Traffic Microsimulation SetupParameter Value
Maximum vehicle speed 14 m/sMaximum vehicle acceleration 2.6 m/s2
For all communications, the complete network stack, in-cluding ARP, is simulated and wireless modules are config-ured to closely resemble IEEE 802.11b network cards trans-mitting at 11 Mbit/s with RTS/CTS disabled. For the simu-
lation of radio wave propagation, a plain free-space model isemployed, with the transmission ranges of all nodes adjustedto a fixed value of 180 m, a trade-off between varying real-world measurements described in related work [8, 26]. Allsimulation parameters used to parameterize the modules ofthe INET Framework are summarized in Table 2.
Table 2: INET Framework Module ParametersParameter Value
6.2 First Simulation ResultsIn order to evaluate the performance of the IVC protocols,
we measured the average speed of the vehicles within ourscenario.
Two simulation scenarios were configured with no IVCtaking place. In the case of free flowing traffic, i.e. simula-tions without any incidents, the speed distribution amongsimulated vehicles is almost homogeneous. Vehicles’ aver-age speed is well below the maximum speed of 14 m/s. Thisis due to cars decelerating at every intersection, which, incombination with high traffic densities on the single, short-est route shared by all vehicles, leads to micro jams. In thesecond case the lead vehicle stopped for a short amount oftime, e.g. due to an accident. Here, the average node speed isreduced by both this stop and by the traffic jam left behind.
Using a traffic incident warning protocol, we expect theroad traffic being influenced by the IVC protocol. As stated
This work is licensed under the CreativeCommons Attribution-ShareAlike 2.0 License.http://creativecommons.org/licenses/by-sa/2.0/
Figure 7: Map of Erlangen, Germany as availablefrom the OpenStreetMap project
before, we examined the effects of two different protocols.Depending on the scale of the simulation, different IVC sce-narios performed differently at helping vehicles avoid theartificially-generated incident.
In order to provide a more detailed look into traffic effectsin this scenario, Figures 6(a) and 6(b) show the effective av-erage speed of vehicles, but present measurements separatedby vehicles’ departure times. Plotted is one single examplerun each, for both the case of free flowing traffic and thecase of IVC with an artificially-triggered incident.
According to the scheduled incident, the lead vehicle isdelayed by 60 s and 240 s, respectively. In the 5×5 scenariodepicted in Figure 6(a), the cars following immediately be-hind are forced into a traffic jam and delayed accordingly.If the IVC message that is sent by the stopped car can bereceived by following cars, they can re-route to a free roadand bypass the jam area. These cars can even drive fasterthan the cars in an incident-free scenario because they donot get delayed at street corners.
Similarly, the incident stopping the leading car involvesall cars following immediately behind it in a traffic jam inthe 16×16 scenario shown in Figure 6(b). This time some ofthem are delayed even further. The first cluster of cars thatwere more than one road away from the incident, however,already had enough time to receive and process the incidentwarning early enough to be able to find alternative routesto the destination, allowing vehicles to reach it even fasterthan they could when they just followed the shortest routein the incident-free scenario. As can be seen, IVC managedto prevent permanent delays on the affected road segment,so even vehicles that were unaware of the incident were ableto continue on their route shortly after the lead vehicle con-tinued its journey: Up to a departure time of just over 240 s,their time spent in the jam linearly decreased towards zero.
Measuring the run-time performance of simulations, weachieved similar results to those obtained in unidirectionally-
coupled simulations [18]. Depending on the exact param-eterization, runs using bidirectionally-coupled simulationstook only insignificantly longer – or, in corner cases, raneven faster – than those using a random waypoint mobilitymodel.
6.3 A More Realistic ExampleBuilding on the first proof-of-concept example in which ve-
hicles traveled on an artificial grid of roads, we now used thecoupled simulation environment to model IVC among traf-fic in the city of Erlangen, Germany. More specifically, wesimulated 200 cars leaving the parking lot of the computerscience building, on average one every 6 s, then heading to abusiness park along an individual, dynamically chosen route.
Serving as the basis for the road layout in this scenariowas map data publicly available from the OpenStreetMapproject. This project unites data from a multitude of freedata sources like the U.S. Census Bureau’s TIGER geo-graphic database, together with community-generated mapdata obtained by volunteers capturing GPS tracks usinghandheld devices, then post-processing these tracks to ob-tain detailed maps. The collected data is available under thepermissive “Creative Commons Attribution-Share Alike” li-cense, allowing the direct modification, as well as free use ofaggregated map data by interested parties.
A rendered representation of the map data, overlaid withthe locations of traffic source and sink nodes, is given inFigure 7. This data modeled the particular section of therequired road network in great detail, accurately reflectingroad attributes such as road type, access restricions, lanecounts, and speed limits. We successfully converted the rawmap data to form a SUMO network, preserving the roadlayout, as well as all pertinent road attributes.
Just like in the previous example, three sets of simulationruns were performed. One set of runs simulated uninhib-ited road traffic. In the second set of runs, a traffic incidentwas simulated by stopping the lead vehicle of cars travellingalong the major artery connecting the university campusand the business park. In the final set of simulation runs, allvehicles were equipped with IVC technology, so stopped ve-hicles could disseminate information about congested roadsegments through a VANET. Vehicles that received suchnotifications could then often completely avoid traffic inci-dents.
Plotted in Figure 8(a) are exemplary results of these threesets of simulation runs, showing the effective average speedof each vehicle in relation to the time it entered the simu-lation. As can be seen, the variance of travel times in thefirst scenario (“free”) was much greater than in the previ-ous example, due to the simulated vehicles now traveling totheir destination along a multitude of different routes. Still,a major portion of the vehicles were involved in the incidenton the major artery that took place in the second scenario,where no IVC took place (“none”). Enabling IVC in the thirdscenario (“udp25”) led to a significant increase of vehicles’speeds, as vehicles that were not too close to the incidentwhen it happened, and thus were caught in the resultingjam, were now able to turn around before they reached theaffected road segment, delaying them only slightly. Othercars managed to avoid the incident, as well as other con-gested road segments, altogether.
The variance of the average speed is also shown in theboxplot in Figure 8(b). In this plot, only individual vehicles
●●●●●●●
●●●
●
●
●●
●●●●●
●●●
●●●●
●
●●●
●●●●●●●●
●
●
●
●●
●
●●
●●●
●●
●
●
●
●●●●
●
●●●
●
●●
●●●
●●●
●●
●
●●●●●●●●
●●●
●●●
●●●
●●●●●
●●
●
●●●●
●
●
●●●●
●
●
●
●●●●●
●●●●●●●●
●●●●●●●●
●●●●●●
●
●
●●
●●●
●
●●●●●
0 200 400 600 800
89
1011
1213
14
Vehicle Departure Time in s
Effe
ctiv
e A
vera
ge S
peed
in m
/s
● freenoneudp25
(a) Average speed ordered by time of departure
●
●●●
●
●●●●●
●●●●●●
free none udp25
89
1011
1213
14
Effe
ctiv
e A
vera
ge S
peed
in m
/s
(b) Average speed grouped by scenario
Figure 8: Average speed of individual vehicles for the free flowing traffic, traffic with an incident, and forUDP-based IVC
starting earlier than 400 seconds are considered to outlinethe characteristics of free flowing traffic, traffic queuing af-ter an incident, and the advantages of UDP-based IVC. Thelatter scenario again outlines the need for bidirectional cou-pling of the simulation tools.
7. CONCLUSIONIn conclusion, it can be said that bidirectionally coupled
road traffic and network simulation provides major advan-tages compared to uncoupled or purely trace-driven simula-tion. The following findings support this observation:
• simple traffic models are inappropriate for road trafficsimulation [27]
• traces of road traffic, either generated using road trafficmicrosimulation or by observing real world traffic, al-low realistic traffic modeling, but IVC protocols cannotbe tested completely as no feedback can be supplied tothe mobility model [13,18]
• using bidirectional coupling, the impact of IVC on roadtraffic can be directly evaluated [19,24]
The simulation framework we developed, Veins (Vehiclesin Network Simulation), provides all necessary functionalityto perform this bidirectional coupling. It relies on state-of-the-art simulators from both domains, thus, it incorporateswell-known models for road traffic microsimulation with acomprehensive selection of models of network protocols. Thepresented proof of concept study demonstrated not only theapplicability but also the need for bidirectional coupling ofroad traffic microsimulation and network simulation.
8. REFERENCES[1] R. Barr, Z. J. Haas, and R. van Renesse. JiST:
Embedding Simulation Time into a Virtual Machine.In EuroSim Congress on Modelling and Simulation,2004.
[2] J. J. Blum, A. Eskandarian, and L. J. Hoffman.Challenges of Intervehicle Ad Hoc Networks. IEEE
Transactions on Intelligent Transportation Systems,5(4):347–351, December 2004.
[3] L. Breslau, D. Estrin, K. Fall, S. Floyd, J. Heidemann,A. Helmy, P. Huang, S. McCanne, K. Varadhan,Y. Xu, and H. Yu. Advances in network simulation.IEEE Computer, 33(5):59–67, May 2000.
[4] E. Brockfeld and P. Wagner. Testing andBenchmarking of Microscopic Traffic Flow SimulationModels. In 10th World conference on transportresearch (WCTR), Istanbul, Turkey, July 2004.
[5] N. Elloumi, H. Haj-Salem, and M. Papageorgiou.METACOR: A macroscopic modeling tool for urbancorridors. Triennal Symposium on TransportationAnalysis, 1:135–150, 1994.
[6] J. Heidemann, N. Bulusu, J. Elson,C. Intanagonwiwat, K.-c. Lan, Y. Xu, W. Ye,D. Estrin, and R. Govindan. Effects of detail inwireless network simulation. In SCS Multiconferenceon Distributed Simulation, pages 3–11, January 2001.
[7] H.-Y. Huang, P.-E. Luo, M. Li, D. Li, X. Li, W. Shu,and M.-Y. Wu. Performance Evaluation of SUVnetWith Real-Time Traffic Data. IEEE Transactions onVehicular Technology (ToVT), 56(6):3381–3396,November 2007.
[8] F. Hui and P. Mohapatra. ExperimentalCharacterization of Multi-hop Communications inVehicular Ad Hoc Network. In 2nd ACM InternationalWorkshop on Vehicular Ad Hoc Networks (VANET2005), Cologne, Germany, September 2005.
[9] M. Killat, F. Schmidt-Eisenlohr, H. Hartenstein,C. Rossel, P. Vortisch, S. Assenmacher, and F. Busch.Enabling Efficient and Accurate Large-ScaleSimulations of VANETs for Vehicular TrafficManagement. In 13th ACM International Conferenceon Mobile Computing and Networking (ACMMobiCom 2007): 4th ACM International Workshop onVehicular Ad hoc Networks (VANET 2007), pages29–38, Montreal, Quebec, Canada, September 2007.
[10] D. Krajzewicz, G. Hertkorn, C. Rossel, andP. Wagner. SUMO (Simulation of Urban MObility);
An open-source traffic simulation. In 4th Middle EastSymposium on Simulation and Modelling(MESM2002), pages 183–187, Sharjah, United ArabEmirates, September 2002.
[11] S. Krauß. Microscopic Modeling of Traffic Flow:Investigation of Collision Free Vehicle Dynamics. PhDthesis, 1998.
[12] D. Li, H. Huang, X. Li, M. Li, and F. Tang. ADistance-Based Directional Broadcast Protocol forUrban Vehicular Ad Hoc Network. InternationalConference on Wireless Communications, Networkingand Mobile Computing 2007 (WiCom 2007), pages1520–1523, 21-25 September 2007.
[13] C. Lochert, A. Barthels, A. Cervantes, M. Mauve, andM. Caliskan. Multiple simulator interlinkingenvironment for IVC. In 2nd ACM InternationalWorkshop on Vehicular Ad hoc Networks (VANET2005), Poster Session, pages 87–88, Cologne,Germany, September 2005. ACM Press.
[14] K. Nagel and M. Schreckenberg. A cellular automatonmodel for freeway traffic. Journal de Physique IFrance, 2:2221–2229, 1992.
[15] M. Piorkowski, M. Raya, A. L. Lugo, M. Grossglauser,and J.-P. Hubaux. Joint Traffic and NetworkSimulator for VANETs. In MICS 2006, PosterSession, Zurich, Switzerland, October 2006.
[16] A. Sobeih, J. C. Hou, L.-C. Kung, N. Li, H. Zhang,W.-P. Chen, H.-Y. Tyan, and H. Lim. J-Sim: asimulation and emulation environment for wirelesssensor networks. IEEE Wireless Communications,13(4):104–119, 2006.
[17] C. Sommer, I. Dietrich, and F. Dressler. A SimulationModel of DYMO for Ad Hoc Routing in OMNeT++.In 1st ACM/ICST International Conference onSimulation Tools and Techniques for Communications,Networks and Systems (SIMUTools 2008): 1stACM/ICST International Workshop on OMNeT++(OMNeT++ 2008), Marseille, France, March 2008.ACM.
[18] C. Sommer and F. Dressler. The DYMO RoutingProtocol in VANET Scenarios. In 66th IEEEVehicular Technology Conference (VTC2007-Fall),pages 16–20, Baltimore, Maryland, USA,September/October 2007. IEEE.
[19] C. Sommer, Z. Yao, R. German, and F. Dressler.Simulating the Influence of IVC on Road Traffic usingBidirectionally Coupled Simulators. In 27th IEEEConference on Computer Communications (IEEEINFOCOM 2008): Mobile Networking for VehicularEnvironments (MOVE 2008), Phoenix, AZ, April2008. IEEE.
[20] N. B. Taylor. The CONTRAM Dynamic TrafficAssignment Model. Networks and Spatial Economics,3:297–322, 2003.
[21] M. Treiber and D. Helbing. RealistischeMikrosimulation von Straßenverkehr mit einemeinfachen Modell. In 16. SymposiumSimulationstechnik (ASIM 2002), Rostock, Germany,September 2002.
[22] M. Treiber, A. Hennecke, and D. Helbing. CongestedTraffic States in Empirical Observations andMicroscopic Simulations. Physical Review E, 62:1805,2000.
[23] A. Varga. The OMNeT++ Discrete Event SimulationSystem. In European Simulation Multiconference(ESM 2001), Prague, Czech Republic, 2001.
[24] S. Y. Wang, C. L. Chou, Y. H. Chiu, Y. S. Tseng,M. S. Hsu, Y. W. Cheng, W. L. Liu, and T. W. Ho.NCTUns 4.0: An Integrated Simulation Platform forVehicular Traffic, Communication, and NetworkResearches. In 1st IEEE International Symposium onWireless Vehicular Communications (WiVec 2007),Baltimore, MD, USA, October 2007. IEEE.
[25] N. Wisitpongphan, F. Bai, P. Mudalige, V. Sadekar,and O. Tonguz. Routing in Sparse Vehicular Ad HocWireless Networks. IEEE Journal on Selected Areas inCommunications (JSAC), 25(8):1538–1556, October2007.
[26] H. Wu, M. Palekar, R. Fujimoto, J. Lee, J. Ko,R. Guensler, and M. Hunter. Vehicular Networks inUrban Transportation Systems. In NationalConference on Digital Government Research(dg.o2005), Atlanta, GA, USA, May 2005.
[27] J. Yoon, M. Liu, and B. Noble. Random waypointconsidered harmful. Twenty-Second Annual JointConference of the IEEE Computer andCommunications Societies (INFOCOM 2003),2:1312–1321, 30 March - 3 April 2003.
