In real world, traffic lights are used to regulate traffic flow moving in different directions.Theexistence oftrafficlightstendsto createa “clustering”effect.In other words, places where there is a traffic light are likely to have a higher node density due to that vehicles are forced to stop at the traffic light to wait for the light to turn green.
Intuitively, a high node density might improve the network connectivity. On the other hand, a higher node density might also suggest a higher chance for packet collision since more nodes might be transmitting at the same time. In addition, the distance between two adjacent traffic lights can have a significant effect on the network connectivity.Specifically,thenetwork can be“fragmented”by thetrafficlightswhen the radio transmission range is smaller than the distance between two adjacent clusters.
In other words, a link breakage can happen when the inter-cluster distance is larger than the radio coverage.
Figure 11 shows the distribution of the number of neighboring nodes when ten traffic lights are included in the simulations. Our results show that each node has twice the number of neighboring nodes when traffic lights are simulated, as compared to the case when traffic lights are not simulated. Here we definea “neighboring node” as the node which is within the radio range of a vehicle. Having a larger number of neighboring nodes typically suggests a better network connectivity.
Figure 11. Clustering effect due to the traffic light
As shown in Figure 12, the packet delivery ratio is improved when the traffic lights are simulated. Note that in this simulation the distance between two adjacent traffic lights is shorter than the given radio range. In addition, we observe that the number of packet collisions increase as we increase the number of traffic sources. As a result, the packet delivery ratio decrease when there are more traffic sources.
Figure 12. Effect of traffic light
To understand the effect of inter-cluster distance on the simulations results, we increase the distance between two adjacent traffic lights (from 200m to 400m) so that the inter-cluster distance is larger than the effective radio distance. As shown in Figure 13, in this scenario we observe frequent link breakage between two adjacent clusters which significant degrade the network performance (in this scenario, the inter-cluster distance is 400m). The effective radio range is around 250m in this experiment.
Figure 13. Effect of inter-traffic-light distance
Finally, we find that the traffic light cycle can also have a significant impact on the network performance. As shown in Figure 14, we observe from simulations that the packet delivery ratio decrease as we increase the traffic light cycle duration (X-axis is the duration of the green light and the red light). While the increased red light cycle increases the cluster size, vehicles are also able to travel farther with a longer green light duration which introduces more link breakage between clusters and results in more packet losses.
Figure 14. Effect of traffic cycle duration B. Driver route choice
In real world, a driver normally has to decide his moving direction at an intersection. He can choose to either go straight, turn left, or turn right. MOVE allows a user to define the turning probability to different directions at each intersection (e.g.
0.5 to turn left,0.3 to go straight and 0.2 to turn left) in the Vehicle Movement Editor.
As shown in Figure 15, we find that different choices of route directions can significantly change the simulation results (the x-y-z notation in Figure 15 means that the car has x% of chance to turn left, y% to go straight and z% to turn right).
Figure 15. Effect of driver route choice
In addition, a driver may choose a path based on different criteria such as travel time, distance, habit, etc. People typically tend to choose the path with the shortest distance to their destinations. However, if everybody all choose the “same”shortest path, it might actually lead to more congestion on the road and longer travel time. On the other hand, a fastest path might not necessarily be the shortest path since a faster path might try to avoid a road segment which is shorter but is more popular.
Intuitively, the choice of a path to the destination could affect nodes density and the network topology as shown in Table I.
Table 1. Comparison of path characteristics between the shortest path and the fastest path
Description The shortest path The fastest travel path
Path choice minimize distance minimum travel time
Node density higher lower
Inter-node distance shorter longer
Probability of having traffic
jam higher lower
As shown in Figure 16, the network performance is better when a popular shortest path is chosen by all the vehicles. While choosing the same shortest path lead road congestion, it also creates a network topology with a higher node density. On the other hand, when choosing a path with the shortest travel time is considered, the vehicles are more uniformly distributed over the whole area, which results in a more sparse network topology. Link breakage is more likely to happen in a sparse network since the inter-node distance is potentially larger. We also observe the packet delivery ratio is increased as we increase the number of nodes in the simulation which effectively increases the network density.
0.00
Figure 16. Evaluation packet delivery ratio in high density network C. Overtaking behavior
In real world, a faster vehicle can overtake some other slower ones when overtaking is allowed on a multi-lane road. Overtaking behavior can have a great effect on the network topology and should be considered. Specifically, when overtaking behavior is not allowed, it usually results in a chain-like topology and a shorter and uniform inter-vehicle distance (the uniform distance is due to that the vehicle needs to maintain a safe distance from the adjacent cars), which often suggests a better network connectivity. We observe a dramatic impact on the network performance when the overtaking behavior is allowed. In addition, we find that the effect of overtaking behavior is less significant when the network density is higher.
As shown in Figure 17, the packet delivery ratios in overtaking-allowed scenario is close to results of no-overtaking scenario when we increase the number of nodes from 250 to 350.
Figure 17. Effect of car overtaking behavior
In summary, we show that details of mobility models such as the existence of traffic lights, driver route choice and car overtaking behavior can have a drastic impact on the VANET simulation results. We argue that the faithfulness of simulation results is proportional to the realism of the parameters and the models used in the simulations. Therefore, selecting appropriate level of details in the mobility model for a VANET simulation is a very important yet challenging task.
5. 相關研究
The details of model could have a critical effect on network simulations. An unrealistic model with insufficient details might produce incorrect results. Heidemann et al. [8] studied how the details of energy and radio propagation models affect the result of sensor network simulations. Zhang et al. [16] used traces taken from UMassDieselNet project [17] to study the effect of mobility models on the performance of DTN. They showed a finer grained route-level model of inter-contact times predict performance much more accurately than the coarser-grained all-bus-pairs aggregated model. This suggests that one must take care in choosing the right level of model granularity when modeling mobility-related measures such as inter-contact times. Complementary to previous studies, in this work we look at the effect of model details on VANET simulations. Our work mainly builds on prior work in MANET mobility models and VANET simulators.
A. Mobility models
Random WayPoint (RWP) [18] is an earlier mobility model widely used in MANET simulation [19], [20], [21], [22]. RWP assumes that nodes can move freely in a simulation area without considering any obstacle. However, in a VANET environment vehicles are typically restricted by streets, traffic lights and obstacles.
Hong et al. [23] proposed a Reference Point Group Mobility (RPGM) model to characterize the relationship between mobile hosts. Bettstetter et al. [24] present a Random Direction Model which introduces a stop-turn-and-go behavior which can mimic the vehicle behavior at the intersections. Camp et al. [25] surveyed different mobility models and divided them into two categories: entity models and group models. Bai et al. [26], [27] did a similar survey and further introduced Freeway and Manhattan mobility models in which car following and overtaking behaviors are included. Saha et al. [7] proposed a macro mobility model based on TIGER map database. This work considers the use of Dijkstra shortest path algorithm to select the path from source to destination. Jardosh et al. [28] present an obstacle mobility model that considers the placement of obstacles in the simulation and discussed that the effect of obstacles on the signal propagation. Stepanov et al. [29] described a spatial model that considered path selection and user movement dynamics (such as road congestion and carfollowing behavior). Japp et al. [30] present a city mobility model that is based IDM (Intelligent-Driver Model). Treiber et al. [31] discussed a model that support car turning at the intersections. Zimmermann et al. [32] proposed a mobility model for urban environments where the paths are computed based on Voronoi graphs and the vehicles movement is constrained by the computed paths.
Street RAndom Waypoint (STRAW) [15] model considered traffic light control and
car following. It uses shortest path algorithm to calculate movement path. Mahajan et al. [33] discussed Stop Sign Model (SSM), Probabilistic Traffic Sign Model (PTSM), and Traffic Light Model (TLM) in the context of a traffic control system. In SSM, every vehicle stops at the stop sign for a fixed duration time. PTSM use a probability p to decide if the vehicle needs to stop at the intersection. In TLM, traffic from different directions are considered for adjusting traffic light cycle to minimize the road congestion. Potnis [34], [33] showed that a simpler SSM model could have significantly different results as compared to a more sophisticated PTSM model.
Marfia et al. [35] employed a similar approach as ours. They used CORSIM [5]
TRANSIMS [36] generate different mobility traces which can be used in Qualnet.
Finally, Baumann et al. [37] proposed two mobility models. One uses vectorized street information from the Swiss Geographic Information System (GIS). The other is based on a microscopic, multi-agent traffic simulator (MMTS) [38] to generate vehicle movement traces. In the GIS-based mobility model, the actual node movement is generated according to the random trip model [39] on the vectorized street map. MMTS models the behavior of people living in the area and the travel plan of each individual as well as road congestion situation are considered in the simulation. This trace-based model, while imitating reality closely, requires a high amount of computing power for generation of traces.
B. VAENT simulators
Groovesim [40] is a topography-accurate street-map based vehicle network simulator and is based on GrooveNet, a geographic routing protocol for vehicular networks. It provides several different modes of operation. In Drive Mode, GrooveSim can process data from a GPS unit to provide a real-time map of the vehicle’scurrentlocation.Itcan also beused asan emulatorinHybrid Simulation Mode where real vehicles on the road and virtual vehicles in the simulation can interact with each other. Groovesim also provides a tool for analyzing the simulation results. One limitation of Groovesim is that it is strongly tied to one specific routing protocol (i.e. GrooveNet), which limits its use for simulating other routing protocols in a VANET environment. In addition, GrooveSim does not provide mobility traces for network simulators.
STRAW is an extension of SWANS (Scalable Wireless Ad Hoc Network Simulator) [41], a Java-based simulator for wireless simulations. STRAW contains simulation tools for generating mobility models and traffic models and is also able to use real street maps like TIGER data to build the road topology. However, currently the mobility models can be supported by STRAW is limited. For example, while STRAW supports multiple lanes, the vehicles are not allowed to change lane and the
starting position is not configurable. Another drawback of this tool is its dependency on SWANS. Finally, STRAW does not provide any GUI that allows the users to visualize the movements of cars.
BonnMotion [42] is a simple tool that can be used to create and analyses mobility scenarios. Similar to MOVE, the mobility scenarios created by BonnMotion can be exported to ns-2 and qualnet. However, BonnMotion only models basic motion constraints and does not consider any micro-mobility. Furthermore, BonnMotion is a text-based application that runs on a command shell and does not provide any graphical user interfaces as MOVE does. Complementary to these previous efforts, our work emphasizes on creating a tool that allows users to rapidly generate realistic mobility models for VANET simulations.
6. 成果發表 6.1. 會議論文
(A) Kun-chan Lan and Chien-Ming Chou,“Realistic Mobility Models for Vehicular Ad hoc Network (VANET) Simulations,”Accepted to appear in ITST2008.
6.2. 期刊論文
(A) Under preparation 6.3. 海報(Poster)
(A) Kun-chan Lan and Chien-Ming Chou, “On the Effects of Detailed Mobility Models in Vehicular Network Simulations,”Accepted to appear in MobiQuitous 2008.
(B) Chien-Ming Chou and Kun-chan Lan, “On the Effects of Detailed Mobility Models in Vehicular Network Simulations,”Accepted to appear in MobiCom 2008.
7. 結論
In this work, we first describe a tool MOVE which is based on an open source micro-traffic simulator SUMO. MOVE allows user to quickly generate realistic mobility models for vehicular network simulations. We show that the details of a mobility model such as the existence of traffic lights, driver route choice and car overtaking behavior can have a significant impact on the simulation results. Care should be taken if simple mobility models are used for evaluation of VANET as the results might not be as close to reality as expected. Our next step is to use MOVE to understand the effect of user travel plan (such as path selection, means of transportation, etc) on VANET simulations. We first evaluated congestion significant effect in VANET. The results depicted congestion can increase network performance.
We have made MOVE publicly available and can be downloaded via the following URL- http://lens1.csie.ncku.edu.tw/MOVE/. In our current implementation, the movements of vehicles are based on static configurations predefined in the Vehicle Movement Editor. In other words, the mobility model is first generated off-line and then used by a network simulator like ns-2. In the next version of our software, because an accident can change mobility model in any time, we plan to build an interface to tightly integrate SUMO and ns-2. Such an interface will allow that vehicle state information (such as location, speed, direction, etc) can be fed into ns-2 in real time. Hence, during the simulation the vehicles can dynamically adjust their routes based on different traffic scenarios and communication techniques employed.
參考資料
[8] J. Heidemann, N. Bulusu, J. Elson, C. Intanagonwiwat, K. chan Lan, Y. Xu, W.
Ye,D.Estrin,and R.Govindan,“Effectsofdetailin wirelessnetwork simulation,”in CNDS 2002, Jan. 2001.
[Online]. Available: http://www.isi.edu/~johnh/PAPERS/Heidemann00d.html
[9]S.S.ofUrban MObility,“http://sumo.sourceforge.net/.”
[10] TIGER (Topologically Integrated GEographic Encoding and Referencing),
“http://www.census.gov/geo/www/tiger/.”
[11]S.Floyd and V.Paxson,“Difficultiesin simulating theInternet,”ACM/IEEE Transactions on Networking, vol. 9, no. 4, pp.
392–403, Feb. 2001. [Online]. Available: http://www.icir.org/vern/papers.html [12]K.tutorial,“http://www.keyhole.com/kml/kmltut.html.” Vehicular Ad Hoc Networks (VANET), Cologne, Germany, Sep. 2005.
[Online]. Available: http://www.aqualab.cs.northwestern.edu/publications/DChoffnes05vanet.html in Mobile Computing, vol. 35, Kluwer Academic, 1996, pp. 153–181.
[19]S.R.Das,C.E.Perkins,and E.M.Royer,“Performancecomparison oftwo
on-demand routing protocolsforad hocnetworks,”in Proc. IEEE Infocom, 2000.
[20]G.Holland and N.H.Vaidya,“Proc.mobicom,acm intern.conf.on mobile computing and networking,”in Proc. IEEE Infocom, Aug. 1999.
[21]C.E.Perkinsand E.M.Royer,“Ad-hoc on-demand distancevectorrouting,”in Proc. IEEE Workshop on Mobile Computing Systems and Applications, Feb. 1999.
[22] P. Johansson, T. Larsson, N. Hedman, B. Mielczarek, and M. Degermark,
“Routing protocols for mobile ad-hoc networks - a comparative performance
Special issue on Mobile Ad Hoc Networking: Research, Trends and Applications, vol.
2, no. 5, 2002, pp. 483–502.
[26]F.Bai,N.Sadagopan,and A.Helmy,“Important:aframework to systematically analyze the impact of mobility on performance of routing protocols for ad hoc networks,”in The 22th IEEE Annual Joint Conference on Computer Communications and Networking INFOCOM 2003, 2003, pp. 825–835.
[27]F.Bai,N.Sadagopan,and A.Helmy,“The importantframework foranalyzing the impact of mobility on performance of routing for ad hoc networks,” AdHoc Networks Journal - Elsevier Science, vol. 1, no. 4, pp. 383–403, Nov. 2003.
[28] A. P. Jardosh, E. M, Belding-Royer,K.C.Almeroth,and S.Suri,“Real-world environmentmodelsformobilenetwork evaluation,”Mar.2005.
[29] I. Stepanov, P.-J.Marron,and K.Rothermel,“Mobility modeling ofoutdoor scenarios for manets,” in Proc. of the 38th annual Symposium on Simulation ANSS ’05,Apr. 2005, pp. 312–322.
[30]M.B.Sven Jaap and L.Wolf,“Evaluation ofrouting protocolsforvehicularad hocnetworksin city trafficscenarios,”in Proc. of the 5th International Conference on Intelligent Transportation Systems Telecommunications (ITST), 2005.
[31]M.Treiber,A.Hennecke,and D.Helbing,“Congested trafficstatesin empirical observationsand microscopicsimulations,”IPhys. Rev., vol. 62, no. 2, Aug. 2000.
[32] H.-M.Zimmermann,I.Gruber,and C.Roman,“A voronoi-based mobility model forurban environments,” in Proc. of the European Wireless 2005 (EW05), Apr. 2005.
[33]A.Mahajan,N.Potnis,K.Gopalan,and A.A.Wang,“Urban mobility models forvanets,”in Proc. of the IEEE Workshop on Next Generation Wireless Networks (WoNGeN), Dec. 2006.
[34] N. Potnis and A. Mahajan, “Mobility models for vehicular ad hoc network simulations,”in proc. of the 44th annual Southeast regional conference, ACM, 2006, pp. 746–747.
[35]G.Marfia,G.Pau,E.Giordano,E.D.Sena,and M.Gerla,“Evaluating vehicle network strategies for downtown portland: Opportunistic infrastructure and importanceofrealisticmobility models,”in proc. of MoBiOpp 2007, co-located with ACM/USENIX International Conference on Mobile Systems, Applications, and Services (MobiSys 2007), Jun. 2007.
[36]TRANSIMS:,“http://transims.tsasa.lanl.gov/.”
[37]R.Baumann,S.Heimlicher,and M.May,“Towardsrealisticmobility modelsfor vehicular ad-hocnetworks,”in 2007 Mobile Networking for Vehicular Environments, May 2007, pp. 73–78.
[38]B.Raney,N.Cetin,A.Vollmy,M.Vrtic,K.Axhausen,and K.Nagel,“Towards a microscopic traffic simulation of all of switzerland,” in ICCS 2002, 2002, pp.
371–380.
[39] J.-Y.L.Boudecand M.Vojnovic,“Perfectsimulation and stationarity ofaclass ofmobility models,”in IEEE Infocom, 2005.
[40] R. Mangharam, D. S. Weller, D. D. Stancil, R. Rajkumar, and J. S. Parikh,
“GrooveSim: A topography-accurate simulator for geographic routing in vehicular network,”in Proc. of the 2nd ACM International Workshop on Vehicular Ad Hoc Networks (VANET), Cologne, Germany, Sep. 2005.
[Online]. Available: http://www.andrew.cmu.edu/