An Architecture Study of AdHoc Vehicle Networks Richard Fujimoto Randall Guensler Hao Wu Michael Hunter Computational Science & Engineering Civil and Environmental Engineering College of Computing College of Engineering Georgia Institute of Technology The Costs of Mobility Safety: 6 Million crashes, 41,000 fatalities in U.S. per year ($150 Billion) Congestion: 3.5 B hours delay, 5.7 B gal. wasted fuel per year in U.S. ($65 Billion) Pollution: > 50% hazardous air pollutants in U.S., up to 90% of the carbon monoxide in urban air > 3M 1M - 3M 0.5 - 1M < 0.5 Million Disproportionate increase in car ownership relative to population growth in China, India Source: 2005 Annual Urban Mobility Report (http://mobility.tamu.edu) Texas Natural Resource Conservation Commission (http://www.tnrcc.state.tx.us/air) Intelligent Transportation Systems www.georgia-navigator.com ITS deployments: Traffic Management Centers (TMC) – – Roadside cameras, sensors, communicate to TMC via private network Disseminate information (web, road signs), dispatch emergency vehicles Infrastructure heavy – – – Expensive to deploy and maintain; limited coverage area Limited traveler information Limited ability to customize services for individual travelers Current Trends Smart Vehicles On-board GPS, digital maps Vehicle, environment sensors Significant computation, storage, communication capability Not power constrained Dedicated Short Range Communications (DSRC) 5.850-5.925 GHz V2V, V2R communication 802.11p protocol 7 channels, dedicated safety channel 6- 27 Mbps Up to 1000 m range U.S. DOT Vehicle Infrastructure Integration (VII) Initiative Public/private partnership “Establishment of vehicle-to-vehicle and vehicle-to-roadside communication capability nationwide” Improve safety, reduce congestion Mobile Distributed Computing Systems on the Road Roadside BaseStation Vehicle-to-vehicle communication Roadside-to-vehicle communication Applications Collision warning/avoidance – – Unseen vehicles Approaching congestions/hazards Traffic/road monitoring Emergency vehicle warning, signal warning Internet Access Traveler & Tourist Assistance Entertainment Objectives Motivating question: Can networks composed of smart vehicles be used to collect and disseminate information in urban / rural transportation systems? • Augment or replace infrastructure deployments Challenges • Create realistic models for mobility by developing, populating, and calibrate simulations specific to data for the Atlanta metropolitan area • Develop simulation modeling tools for traffic, vehicleto-vehicle, and vehicle-to-roadside communications to support the development and evaluation of future generation intelligent transportation systems • Evaluate the performance limits of multi-hop vehicleto-vehicle communication for realistic test conditions Spatial Propagation Problem Message Spatial Propagation Problem: How fast can information propagate with vehicle forwarding? Focus on V2V ad hoc networks (802.11) in order to understand the limitations of message forwarding Observations One dimensional partitioned network Vehicle movement helps propagate information Vehicle Ad Hoc Networks H Time t0 Time t1 Time t2 Informed Vehicle Uninformed Vehicle Partition Partitioned Network Forward mode Message forwarding within a partition Catch-up mode Distance Time t3 Vehicle movement allows message propagation between partitions Cyclic Process Forward Mode Catch-up Mode Time Time-space Trajectory Analytic Models H. Wu, R. M. Fujimoto, G. Riley, “Analytical Models for Information Propagation in Vehicle-to-Vehicle Networks,” IEEE Vehicular Technology Conference, September 2004. A single road with one way traffic Vehicle movement follows undisturbed traffic model Distance for 100s (m) 16000 Sparse Netow ork Model Dense Netw ork Model Simulation 14000 12000 10000 8000 6000 4000 2000 1.25 1.1 0.95 0.8 0.65 0.5 0.35 0.2 0.05 0 Flow Rate (vehicles/s) Sparse network model -- Small partition size – – Information propagation principally relies on vehicle movement. Message propagation speed approaches maximum vehicle speed. Dense network model -- Large partition size – – Independent cycles Renewal reward process Reward: message propagation distance during each cycle Integrated Distributed Simulations CORSIM Microscopic traffic simulation Vehicle-tovehicle and vehicle-toinfrastructure wireless communication Distributed simulation over LANs and WANs QualNet Traffic Simulator Comm. Simulator Run Time Infrastructure Software Federation management Pub/Sub Communication LAN/Internet Synchronization (Time Management) Traffic Simulation Model (Guensler, Hunter, et al.) • One-foot resolution United States Geological Survey (USGS) orthoimagery aerial photos used to code lanes, turn bay configurations, and turn bay lengths for each intersection • Traffic volumes, signal control plans, geometric data, speed limits, etc., obtained from local transportation agencies Traffic Corridor Study Area I-75 and surrounding arterials in NW Atlanta 189 nodes (117 arterial, 72 freeway) 45 signalized nodes 365 one-way links (295 arterial, 70 freeway) 101.4 arterial miles 16.3 freeway miles (13.6 mainline, 2.7 ramp) Model Calibration & Validation H. Wu, J. Lee, M. Hunter, R. M. Fujimoto, R. L. Guensler, J. Ko, “Simulated Vehicle-to-Vehicle Message Propagation Efficiency on Atlanta’s I-75 Corridor,” Journal of the Transportation Research Board, 2005. Speed(MPH) Comparison of Measured and Simulated Vehicle Speeds 70 Measured 60 Simulated 50 40 30 20 10 R os ew ry el lR R d. d. r. Fe r Pa ce s or th sid e ar a t. P M D n d. R M il l Bl vd . oo re N W Arterials Anomalous (simulated) delays observed at some locations – M d. ar ie t ta R M Fe rry ef oo rs ol l ie rR d. D C I-7 5 So ha t ta ut h ch oo ch ee Av e C I-7 5 N or th 0 Field surveys completed at six intersections to calibrate model Validation using instrumented vehicle fleet collecting second-bysecond speed and acceleration data – – – GPS data from 7 AM to 8 AM peak used 591 weekday highway trips (Feb.-May 2003) 601 weekday highway trips (July-Sept. 2003) Mobility-Centric Data Dissemination for Vehicle Networks (MDDV) H. Wu, R. M. Fujimoto, R. Guensler, M. Hunter, “MDDV: Mobility-Centric Data Dissemination Algorithm for Vehicular Networks,” ACM Workshop on Vehicular Ad Hoc Networks (VANET), October 2004. No end-to-end connection assumption – – – Opportunistic forwarding [Fall, SIGCOMM 2003] Trajectory-based forwarding [Niculescu & Nath, Mobicom’03] Geographic forwarding [Mauve, IEEE Networks 15 (6)] Compute trajectory to destination region Group forwarding: Set of vehicles holding message “closest” to destination region actively forward message toward destination Group membership – – – Vehicle stores last known location/time of message head candidate; forwards information with message Join group if (1) moving toward destination along trajectory and (2) reach estimated head location (or closer) less than Tl time units after head Leave group if (1) leave trajectory or (2) receives same message indicating head is closer to the destination Propagation Delay (simulation) Delay (s) H. Wu, J. Lee, M. Hunter, R. M. Fujimoto, R. L. Guensler, J. Ko, “Simulated Vehicle-to-Vehicle Message Propagation Efficiency on Atlanta’s I-75 Corridor,” Journal of the Transportation Research Board, 2005. End-to-End Propagation Delay 120 EVENING PEAK 100 80 60 40 20 0 0.05 0.1 0.15 0.2 0.25 Penetration Ratio Penet r at i o n R at i o Delay to propagate message 6 miles southbound on I-75 Relatively heavy traffic conditions Penetration ratio: fraction of instrumented vehicles End-to-End Delay Distribution Delay to propagate message 6 miles along I-75 (southbound) Heavy (evening peak) and light (nighttime) traffic Penetration ratio: fraction of instrumented vehicles Significant fraction of messages experience a large delay Delivery Ratio Mobility-centric Data Dissemintation for Vehicle Networks (MDDV) Delivery Rat io 1.2 1 0.8 0.6 0.4 0.2 0 0.1 0.2 0.3 0.4 0.5 Delay (s) P e ne t r a t i on Ra t i o 500 Avg Delay M ax Delay M in Delay 0. 5 400 300 200 100 0 0. 1 0. 2 0. 3 0. 4 Penetration Ratio P e ne t r a t i on R a t i o MDDV: opportunistic forwarding algorithm Morning rush hour traffic Propagate information to destination 4 miles away Delivery ratio: fraction delivered before expiration time (480 seconds) Large variation in delay observed Field Experiments: Goals Characterize communication performance in a realistic vehicular environment Identify factors affecting communication Lay the groundwork of realizing communication services Demonstrate and assess the benefits of multi-hop forwarding When the Rubber Meets the Road H. Wu, M. Palekar, R. M. Fujimoto, R. Guensler, M. Hunter, J. Lee, J. Ko, “An Empirical Study of Short Range Communications for Vehicles,” ACM Workshop on Vehicular Ad Hoc Networks (VANET), September 2005. In-vehicle systems – – – – Roadside station using the same equipment Software – – Iperf w/ GPS readings Data forwarding module Location – Laptop GPS receiver 802.11b wireless card External antenna I-75 in northwest Atlanta, between exits 250 and 255 Un-congested traffic Clear weather Vehicle-to-Roadside (V2R) Communication Exit 255 -2440m -2270m -1400m Exit 254 -700m Peachtree Battle Bridge 0 600m Trees Exit 252B N S Exit 252A 1000m 2500m 3000m 3370m Exit 250 4250m 4700m 4570m V2R Performance Success Ratio - Percentage of packet transmissions received by the receiver North South Northbound Southbound Success Ratio View facing north Bridge 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% -500 0 -1500 -1000 View facing south 500 1000 Trees Peachtree Battle Trees Exit 254 north 1500 2000 Distance (m) Vehicle-to-Vehicle (V2V) Communication Exit 255 -2440m -2270m -1400m Exit 254 -700m Peachtree Battle Bridge 0 600m Trees Exit 252B N S Exit 252A 1000m 2500m 3000m 3370m Exit 250 4250m 4700m 4570m V2V Performance (Southbound) North v2v Distance = 150m v2v Distance = 300m Bridge South 100% 80% Success Ratio Exit 255 -2440m -2270m 60% -1400m 40% 20% -700m 0% -3000 -2000 -1000 0 1000 2000 3000 4000 0 5000 Exit 254 Peachtree Battle Bridge Sender Location (m ) 600m Trees v2v Distance = 400m v2v Distance = 700m Success Ration North -3000 -2000 Bridge 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% -1000 0 South Exit 252B N S 1000m Exit 252A 2500m 3000m Exit 250 3370m 4250m 4700m 4570m 1000 2000 3000 4000 5000 Sender Location (m) Multi-hop Communication Exit 255 -2440m -2270m -1400m Exit 254 -700m Peachtree Battle Bridge 0 600m Trees Exit 252B N S Exit 252A 1000m 2500m 3000m 3370m Exit 250 4250m 4700m 4570m Performance Comparison Single hop, northbound Single hop, southbound Tw o hops, southbound, v2v distance = 340m Tw o hops, southbound, v2v distance = 180m Tw o hops, northbound, v2v distance = 320m Tw o hops, northbound, v2v distance = 460m 80% 60% 40% 20% -1200 -700 0% -200 Success Ratio 100% 300 800 1300 1800 Distance (m ) Summary Mobile distributed computing systems on the road are coming – – – Simulation methodology is essential to design vehicle networks, e.g., to determine a necessary penetration ratio for effective communication – – Safety likely to be the initial primary application System monitoring also early application Enable wide variety of commercial applications Realistic evaluation of vehicular networks requires careful consideration of mobility Federating simulation models can play a key role Vehicle-to-vehicle communication can be used to propagate information for applications that can tolerate some data loss and/or unpredictable delays – – V2V communication provides a means to supplement infrastructurebased communications Must weigh benefits against implementation complexity Future Directions Architectures of the future will likely include a mix of technologies – – WWAN, WLAN (e.g., DSRC), V2V Roadside computing stations, Internet gateways Transition from data draught to data flood will create new technical challenges – – – Management of bandwidth Management of computing resources; vehicle grids Data challenges: cleaning, aggregating, mining Wireless Infrastructure Technologies Wireless Technologies (in order of decreasing coverage) – Wireless Wide Area Networks (WWAN) – Wireless Metro Area Networks (WMAN) – IEEE802.11x (T-mobile hop spots) High bandwidth: 802.11b provides 11 Mbps, 802.11 a/g offers 54 Mbps Low coverage (250m) Wireless Personal Area Networks (WPAN) Fixed broadband wireless link (WiMAX -- IEEE 802.16) Wireless Local Area Networks (WLAN) – Cellular networks (2nd Generation, 2.5G, 3G, 4G) High coverage (up to 20 km) Low bandwidth: Verizon BroadbandAccess provides up to 2 Mbps upstream, the Cingular Edge provides up to 170 Kbps upstream Bluetooth Larger coverage -> Increased cost, low bandwidth Network Architecture Options Backbone WWAN BS WLAN AP WWAN last hop: broad coverage, limited capacity WLAN access points to improve capacity – – WLAN AP In addition to, or rather than WWAN Many access points vs. coverage tradeoff Add WLAN multihop (vehicle-to-vehicle) communication – – Extend WLAN coverage, reduce number of access points Requires presence of vehicles Required WWAN Capacity vehicle data rate = 16Kbps (a modest value) 7 WLAN access points (for hybrid architectures) (Kbps) Required WWAN Capacity WWAN WWAN WWAN WWAN 35000 30000 25000 last-hop last-hop + WLAN last-hop last-hop + 2-hop WLAN last-hop + 3-hop WLAN 28.8 Mbps 20000 15000 10000 5.6 Mbps 5000 0 0 Not Linear 0.2 0.4 0.6 0.8 1 Penetration Ratio WWAN does not scale well. A hybrid architecture can increase the system capacity and reduce the WWAN data traffic load. Required WLAN Access Points to Provide Sufficient Capacity vehicle data rate = 16 Kbps, road length = 11,000 m, number of instrumented vehicles = 1800 * penetration ratio, aggregated WWAN data rate = 6 Mbps WLAN last-hop WWAN last-hop + WLAN last-hop WWAN last-hop + 2-hop WLAN WWAN last-hop + 3-hop WLAN # of Access Points 25 20 15 10 5 0 0 0.2 0.4 0.6 0.8 1 Penetration Ratio Fixed number required for WLAN last-hop architecture Hybrid architecture can greatly reduce the number Multi-hop forwarding can reduce number further WLAN Coverage Range Coverage range: expected length of road segment within which vehicles can access a WLAN access point using at most m hops hops = 1 analysis, hop limit = 2 analysis, hop limit = 3 simulation, hop limit = 2 simulation, hop limit = 3 Normalized WLAN Coverage 3.5 3 2.5 2 1.5 1 0.5 0 0 0.025 0.2 0.4 0.6 0.8 1 Penetration Ratio Instrumented vehicles will likely be sufficiently dense Further coverage increase minor when instrumented vehicle density reaches a saturation value (penetration ratio 0.3 above) Design Implication Vehicular network design requires: – – Careful assessment of cost / performance tradeoffs Addressing changing vehicle traffic conditions Multi-hop forwarding – – – – Pro: extend coverage -> reduce number of access points > reduce cost Con: reduced channel capacity, additional system complexity (routing, billing & security) Questionable except in places with cost or other constrains Voluntary cooperation is beneficial in improving communication performance Design Implication (Cont.) Continuous connectivity – – – Intermittent connectivity – – – WWAN: does not scale well. WLAN last-hop: simple, easy deployment, provide high throughput, require a large number of access points WWAN + WLAN: increase system capacity WLAN-based solution Whether to allow multi-hop forwarding is governed by a tradeoff between cost and system complexity. Connectivity probability in every location can be estimated using our models. Deal with vehicle traffic dynamics – – Overprovision (for hard-to-predict variations) Adaptation (for predictable variations) References H. Wu, J. Lee, M. Hunter, R. M. Fujimoto, R. L. Guensler, J. Ko, “Simulated Vehicle-to-Vehicle Message Propagation Efficiency on Atlanta’s I-75 Corridor,” Journal of the Transportation Research Board, 2005. H. Wu, R. M. Fujimoto, R. Guensler, M. Hunter, “An Architecture Study of Infrastructure-Based Vehicular Networks,” Eighth ACM/IEEE International Symposium on Modeling, Analysis, and Simulation of Wireless and Mobile Systems,” October 2005. R. M. Fujimoto, H. Wu, R. Guensler, M. Hunter, “Evaluating Vehicular Networks: Analysis, Simulation, and Field Experiments,” Cooperative Research in Science and Technology (COST) Symposium on Modeling and Simulation in Telecommunications, September 2005. H. Wu, M. Palekar, R. M. Fujimoto, R. Guensler, M. Hunter, J. Lee, J. Ko, “An Empirical Study of Short Range Communications for Vehicles,” ACM Workshop on Vehicular Ad Hoc Networks (VANET), September 2005. Lee, J., M. Hunter, J. Ko, R. Guensler, and H.K. Kim, "Large-Scale Microscopic Simulation Model Development Utilizing Macroscopic Travel Demand Model Data", Proceedings of the 6th Annual Conference of the Canadian Society of Civil Engineers, Toronto, Ontario, Canada, June 2005. H. Wu, M. Palekar, R. M. Fujimoto, J. Lee, J. Ko, R. Guensler, M. Hunter, “Vehicular Networks in Urban Transportation Systems,” National Conference on Digital Government Research, May 2005 H. Wu, R. M. Fujimoto, R. Guensler, M. Hunter, “MDDV: Mobility-Centric Data Dissemination Algorithm for Vehicular Networks,” ACM Workshop on Vehicular Ad Hoc Networks (VANET), October 2004. H. Wu, R. M. Fujimoto, G. Riley, “Analytical Models for Information Propagation in Vehicle-to-Vehicle Networks,” IEEE Vehicular Technology Conference, September 2004. B. Fitzgibbons, R. M. Fujimoto, R. Guensler, M. Hunter, A. Park, H. Wu, “Simulation-Based Operations Planning for Regional Transportation Systems,” National Conference on Digital Government Research, pp. 175-176, May 2004. B. Fitzgibbons, R. M. Fujimoto, R. Guensler, M. Hunter, A. Park, H. Wu, “Distributed Simulation Testbed for Intelligent Transportation System Design and Analysis,” National Conference on Digital Government Research, pp. 308-309, May 2004.