1. UCLA ENGINEERING Computer Science RobustGeo: a Disruption-Tolerant Geo-routing Protocol Ruolin Fan, Yu-Ting Yu *, Mario Gerla UCLA, Los Angeles, CA, USA {ruolinfan, gerla}@cs.ucla.edu * Qualcomm Research, Bridgewater, NJ, USA yutingy@qti.qualcomm.com

2. UCLA ENGINEERING Computer Science INTRODUCTION

3. UCLA ENGINEERING Computer Science Location-Based Routing A class of routing algorithms that use locations of forwarding nodes to route packets Scalable: no need for each node to know overall network topology Promising routing method for VANETs – Mercurial network topology – Limited bandwidth

4. UCLA ENGINEERING Computer Science Location-Based Routing Greedy Forwarding – Use geographical locations – Find neighboring nodes closest to the destination geographically – Forward the packet to that node S D

5. UCLA ENGINEERING Computer Science Location Based Routing Perimeter Forwarding – Sometimes no immediate closer node to the destination can be found Local maximum encountered – Route around the perimeter S D L.M

6. UCLA ENGINEERING Computer Science Issues with Location Based Routing Works well for mobile scenarios: VANET routing But, cannot handle temporary topology disruptions: – Complete disconnection, ie partitioned network with no end-to-end routes Cannot be dealt with by traditional geo-routing Perimeter routing fails if net is partitioned – Can be short lived, yet frequent, eg vehicular traffic patterns caused by traffic lights

7. UCLA ENGINEERING Computer Science Delay-Tolerant Networks (DTN) Originally proposed for networks that experience long disconnections (eg space) Exploit mobility of nodes to physically carry data from the source to the destination (carry- and-forward) DTN delivers data in intermittent networks (eg space), but latency is unacceptable in VANETs Insight: use Replication to alleviate latency

8. UCLA ENGINEERING Computer Science RobustGeo Combines Geo-routing with DTN routing Greedy-forwarding when possible When local-maximum is hit: – Perimeter forwarding + DTN routing – Controlled packet replication to increase delivery probability

9. UCLA ENGINEERING Computer Science SYSTEM DESIGN

10. UCLA ENGINEERING Computer Science Design Overview Geo-routing component – Used under all normal greedy-forwarding scenarios – Can make use of all existing geo-forwarding techniques based on that of GPSR Disruption-tolerance component – Used for resolving local maximum scenarios – Perimeter route a replica of the original packet in case the local maximum is temporary – Broadcast the original packet if intermittency lasts too long

11. UCLA ENGINEERING Computer Science Perimeter Routing Original packet saved into a buffer called the DTQ (delay-tolerant queue) Perimeter forwarding for the packet replica using the right-hand rule until: 1.Greedy route found, ack returned, DTQ dropped 2.TTL exceeded, replicated packet dropped If both the original packet and its replica finds a greedy route, one of them is dropped

12. UCLA ENGINEERING Computer Science Perimeter + Broadcast duplicates

13. UCLA ENGINEERING Computer Science Packet Broadcasting Packets in the DTQ with no previous broadcast are broadcast periodically (every 6 sec) – Explore multiple paths to increase delivery rate – Single-hop broadcasting Node that receives a packet via broadcasting saves it in DTQ and waits for greedy path Broadcasting period of 6 s fits vehicular traffic

14. UCLA ENGINEERING Computer Science DTQ Scheduling

15. UCLA ENGINEERING Computer Science ANALYSIS: A LIMIT ON PACKET REPLICATION

16. UCLA ENGINEERING Computer Science Replication Mechanism Over-replication of packets can congest the network Packets are replicated only once in a local maximum situation – A single packet replica is generated when perimeter routing is attempted – Long recovery time can generate multiple packet replicas due to periodic broadcasting A packet replica originating from a broadcast cannot generate more replicas via broadcasting

17. UCLA ENGINEERING Computer Science Mathematical Model The total number of replicated packets generated due to intermittencies in the network rep: # replicated packets m: # packets that experience local maximum recovery n: # neighbors that receive the broadcasted packet t: Average local max recovery time π: Broadcasting period of each node P b : Probability a broadcast packet found a path P p Probability a perimeter-routed packet found a path K: Total number of intermittencies

18. UCLA ENGINEERING Computer Science Graphical Representation for One Packet (m = 1) π = 6; n = 10; P p =0.01; P b = 0.2 Intermittency length t is modeled as a Poisson RV with λ=3

19. UCLA ENGINEERING Computer Science PERFORMANCE EVALUATION

20. UCLA ENGINEERING Computer Science Simulation Parameters NS3 simulator Radio channel: OFDM 36Mbps Propagation Loss Model: Friis WiFi Standard: 802.11a Wifi Mac Type: Adhoc Wifi Mac Transport Protocol: UDP Application: CBR Client-Server Pair (20Kbps) Simulation Time: 200s

21. UCLA ENGINEERING Computer Science Comparison set Pure geo-routing: – simply drops packet that it cannot immediately route, Geo-routing with a DTQ: – improves delay tolerance but without packet brcst, replication; similar to GeoDTN+Nav, Geo-routing with controlled flooding, – DTQ can only pass the packet to 5 unique neighbors before dropping (like Epidemic Routing )

22. UCLA ENGINEERING Computer Science Simple scenario 3 fixed nodes; 2 vehicle clusters (8 nodes)

23. UCLA ENGINEERING Computer Science Simulated Traffic in DC

24. UCLA ENGINEERING Computer Science Simulated Traffic in DC – Delivery Ratio

25. UCLA ENGINEERING Computer Science San Francisco Taxi Traces

26. UCLA ENGINEERING Computer Science Packet Replication: SF Taxi # Intermittencies # Packets 11434 21063 3981 4214 53 65

27. UCLA ENGINEERING Computer Science CONCLUSION

28. UCLA ENGINEERING Computer Science Conclusion We designed RobustGeo to withstand network intermittencies common to urban VANETs Takes advantage of both the store-and- forward and replication strategies of DTNs A viable tradeoff between increasing network overhead and packet delivery ratio A hybrid solution that allows for connections in both reliable and intermittent networks (typical VANET attributes)

29. UCLA ENGINEERING Computer Science Thank You! Questions?