1. Geographic Routing in Vehicular Ad Hoc Networks (VANETS) Kevin C. Lee Computer Science Department University of California, Los Angeles Chair – Professor Mario Gerla

2. 2 Outline  Overview of geographic routing  Summary of previous work  Present LOUVRE Histogram-based density estimation approach  Report GeoDTN+Nav new results

3. 3 Greedy Mode  Nodes learn 1-hop neighbors’ positions from beaconing  A node forwards packets to its neighbor closest to D  Greedy traversal not always possible! x is a local maximum to D; w and y are further from D

4.  Face traversal by right-hand rule  Face change Walking sequence: F 1 -> F 2 -> F 3 -> F 4 Recovery/Perimeter Mode x y z S D F1F1 F2F2 F3F3 F4F4 4 A B C D E I1 I2 I3

5.  Face traversal requires planar graph: cross edges result in routing loops  GG and RNG planarization algorithms  Their disadvantages  Planarization overhead  High hop count  Unit disk assumption, GPS accuracy, etc Planarization 5

6. 6 Outline  Overview of geographic routing  Summary of previous work  Present LOUVRE Histogram-based density estimation approach  Report GeoDTN+Nav new results

7. 7 TO-GO[1, 2] Perimeter forwarding using greedy forwarding Packet skipping a junction node if not changing direction  Eliminate planarization overhead – Roads naturally formed a “planar” graph  Improve routing efficiency – Packets stop @ the junction only when necessary (aka junction lookahead)  Improve packet delivery – Opportunistic forwarding whenever possible Opportunistic routing toward the target

8. 8 GeoCross[3] Routing loop!!  Motivation: Empty intersection -> routing loop -> low packet delivery

9. 9 GeoCross Basic Operations S, R1, [R1R2], R2, B, R3, C, R4, D, R5, [R5R6], R6, E, R7, F, R8, B => No cross link, continue forwarding S, R1, [R1R2], R2, B, R3, C, R4, D, R5, [R5R6], R6, E, R7, F, R8, B, R2, [R2R1], R1, S UR: [R5R6], continue existing loop Can’t forward b/c UR: [R5R6] Packet reaches destination

10. 10 LOUVRE[4]  Recovery mode often expensive; backtracking takes too many steps  Use P2P density information to guide packet routing  LOUVRE: end-to-end routing solution that eliminates recovery forwarding completely D S ? Road 1 ss Density > Thresh = 3 2 3 3 3 5 3 3 00 50 0 3 3 s Overlay routes

11. 11 Limitations & Previous Work  TO-GO:  No planarizaton overhead by taking roads that naturally formed a planar graph  Improve efficiency by junction-lookahead  Opportunistic forwarding to improve packet delivery  GeoCross: Takes care of loop-inducing cross links  LOUVRE: Peer-to-peer density estimation to avoid dead ends and backtracking

12. 12 Outline  Overview of geographic routing  Summary of previous work  Present LOUVRE Histogram-based density estimation approach  Report GeoDTN+Nav new results

13. 13 Drawback of the LOURVRE’S P2P Density Estimation Scheme  Not scalable  The memory overhead increases with the number of nodes  Not accurate  Density does not correlate well with connectivity when it is not uniform NOT CONNECTED

14. Histogram-Based Density Discovery Algorithm[5]  Break up the roads into segments  Nodes within a segment keep track of unique # of cars they have seen in P2P fashion  Nodes receive broadcast beacons to update segment densities in the other segments  Road is connected if 14 120012?0 1200 1210 1210 Segment center 1200 1210 Segment 1 Segment 2 Segment 3 Segment 4 ABC D

15. Advantages of Histogram-Based Approach  Scalable  E.g. 1500-meter road, 250-meter segment length  Only need 6 integers for 6 segments (1500/250)  P2P can only store 6 cars, not enough  More accurate  Each segment size is smaller than the road length  Connectivity correlates better with segment density than road density 15 NOT CONNECTED

16. Evaluation  Connectivity accuracy between P2P and histogram-based approach  Road Percentage Connectivity (RPC) vs. Connectivity Accuracy (CA)  If road is connected, CA = RPC  If road is not, CA = 1 – RPC  Broadcast overhead between P2P and histogram-based approach  1,000 realistic mobility traces 16

17. Connectivity Accuracy between P2P and Histogram 17  P2P underperforms when density is low  This is due to the clustering behavior at two ends of a road

18. Broadcast Overhead between P2P and Histogram  P2P has scalability issue as it needs to keep track of unique cars 18

19. 19 Outline  Overview of geographic routing  Summary of previous work  Present LOUVRE Histogram-based density estimation approach  Report GeoDTN+Nav new results

20. 20 GeoDTN+Nav Motivation [6,7]  Current geographic routing protocols assume connected networks  Connectivity not always guaranteed  Intermittent connectivity possible:  Low vehicle density  Obstacles  Temporal evolving traffic pattern

21.  Basic idea: Exploit mobility to help deliver packets across disconnected networks  The problem now is which node to choose?  Blind random choice:  Might not help  Nodes may move even farther away from the destination  Informed choice:  Better decision  HOW? – WHAT IF we know more about nodes (such as their destination or path information) 21 Which Node?

22.  Harvest neighbors’ dest/path information  Assumption:  Every vehicle has a navigation system  Is it true?  Relaxed Assumption  “Pseudo/Virtual” navigation system 22 Navigation System Helps!

23.  A lightweight wrapper interface interacts with data sources  Provide two unified information:  Nav Info  Destination  Path  Direction  Confidence  0% (Unreliable) ~ 100% (Reliable) 23 Virtual Navigation Interface

24. 24 VNI Example Bus VNI : (Path, 100%) Taxi VNI : (Dest, 100%) w/ Navigation VNI : (Path, 55%) w/o Navigation VNI : (?, 0%)

25.  Introduce third forwarding mode in geo- routing  DTN recovery mode  Complement conventional two-mode geo- routing  Three routing modes  Greedy  Perimeter  DTN 25 GeoDTN+Nav Modes

26.  In recovery mode  Current nodeC  Neighbors N i ( i =1~n)  Hops h  Compute a “switch score” for each neighbor with  Scoring function S  Switch threshold S thresh 26 DTN Mode RULE: If S(C) > S thresh and there exists N i, such that S(N i ) > S thresh and S(N i ) > S(N j ), i ≠ j for all j Switch to DTN mode Forward the packet to N i

27.  S(Ni) = αP(h) + βQ(Ni) + γDir(Ni) where α + β + γ = 1  S(Ni):“Switch score” of Ni  P(h):(0 ~ 1) Partition probability  Q(Ni): (0 ~ 1) Quality of the “mule”  Dir(Ni):(0 ~ 1) Direction of the “mule” towards the dest  P(h) ↑ S(Ni) ↑  If the network is highly suspected to be disconnected, it would be better to switch to DTN  Q(Ni) ↑ S(Ni) ↑  If there is a neighbor which has higher guarantee of delivery of packets to the destination, Q(Ni) would increase S(Ni)  Dir(Ni) ↑ S(Ni) ↑  If the neighbor is heading toward the destination, Dir(Ni) would increase S(Ni)  Q(Ni) and Dir(Ni) functions depend largely on info from VNI!! 27 Scoring Function

28. 28 P(h)  Suspect network connectivity by “traversed hop counts”  RED-like probability function  h min  h max

29. 29 Q(Ni)  Calculate Ni’s “Delivery Quality”  Navigation information  Confidence D1D1 D2D2 D3D3

30. 30 Dir(Ni)  Determine Ni’s “routability”: Can Ni carry the packets?  Ni’s direction wrt destination  Current node’s direction wrt destination Dir(N2) > Dir(N1)

31.  Let  α = β = 0.5, γ = 0  S thresh = 0.5 31 Example: Perimeter to DTN Q(N1) = 0.1 D(N1) = 0.8 S(N1) = 0.25 P(9) = 0.5 Q(B) = 0.5 D(B) = 1 S(B) = 0.50 Q(N2) = 0 D(N2) = 0.2 S(N2) = 0.25 P(8) = 0.4 Q(A) = 0.4 D(A) = 0.2 S(A) = 0.4 Q(N3) = 0.6 D(N3) = 0.5 S(N3) = 0.5 Q(N1) = 0.2 D(N1) = 0.3 S(N1) = 0.35 Q(N2) = 0.7 D(N2) = 0.8 S(N2) = 0.60 Q(N3) = 0.6 D(N3) = 0.9 S(N3) = 0.55

32.  Switch to greedy only if neighbor score is lower AND it’s closer than the node that first entered into DTN 32 Example: DTN to Greedy A Y B X K J D C S(X) = 0.2 S(X) = 0.4 S(B) = 0.6 S(A) = 0.5 S(K) = 0.4 S(J) = 0.3 S(C) = 0.3 S(B) = 0.5 A

33.  Topology: 1500m by 4000m Oakland map from TIGER database  Mobility:  VanetMobisim (100 cars)  50 buses and taxis for mules  Routing protocols: GPCR, RandDTN 33 GeoDTN+Nav Evaluation  Metrics: PDR, hop count, latency

34.  GeoDTN+Nav maintains high PDR because packets are carried mostly by Bus nodes  GeoDTN+Nav beats RandDTN 34 PDR

35.  GeoDTN+Nav latency lower than RandDTN because of its hybrid nature  GPCR latency is low => packets are delivered when network is connected 35 Latency

36.  GeoDTN+Nav higher hop count than RandDTN  Trading high count for PDR and low latency 36 Hop Count

37.  % of Bus nodes and taxi nodes as mules  As the number of bus node increases, PDR increases => bus has better packet delivery  GeoDTN+Nav able to use both types of vehicles provided by VNI 37 GeoDTN+Nav Forwarding Diversity

38. 38 Conclusion  Geographic routing is feasible in VANETs  Yet it is inefficient in a VANET environment  We identified problems of geographic routing in VANETs and propose solutions:  Planarization overhead, routing inefficiency, and signal interference (TO-GO)  Routing loops caused by empty junction nodes (GeoCross)  Expensive recovery (LOUVRE)  Intermittent connectivity (GeoDTN+Nav)

39. 39 Publication 1."Enhanced Perimeter Routing for Geographic Forwarding Protocols in Urban Vehicular Scenarios,“ Kevin C. Lee, Jerome Haerri, Uichin Lee, Mario Gerla, Autonet'07, Washington, D.C., November, 2007. 2."TO-GO: TOpology-assist Geo-Oppertunistic Routing in Urban Vehicular Grids," Kevin C. Lee, Uichin Lee, Mario Gerla, WONS 2009, Snowbird, Utah, February, 2009. 3."GeoCross: A Geographic Routing Protocol in the Presence of Loops in Urban Scenarios," Kevin C. Lee, Pei-Chun Cheng, Mario Gerla, Ad Hoc Networks: January, 2010. 4."LOUVRE: Landmark Overlays for Urban Vehicular Routing Environments," Kevin C. Lee, Michael Le, Jerome Haerri, Mario Gerla, WiVeC 2008, Calgary, Canada, September, 2008. 5."Histogram-Based Density Discovery in Establishing Road Connectivity," Kevin C. Lee, Jiajie Zhu, Jih-Chung Fan, Mario Gerla, VNC, Tokyo, Japan, October, 2009. 6."GeoDTN+Nav: A Hybrid Geographic and DTN Routing with Navigation Assistance in Urban Vehicular Networ," Pei-Chun Cheng, Jui-Ting Weng, Lung-Chih Tung, Kevin C. Lee, Mario Gerla, Jerome Haerri, MobiQuitous/ISVCS 2008, Trinity College Dublin, Ireland, July, 2008. 7."GeoDTN+Nav: Geographic DTN Routing with Navigator Prediction for Urban Vehicular Environments," Pei-Chun Cheng, Kevin C. Lee, Mario Gerla, Jérôme Härri, Mobile Networks and Applications: Volume 15, Issue 1 (2010), Page 61.