Improving Geographical Routing for Wireless Networks with an Efficient Path Pruning Algorithm

1. Improving Geographical Routing for Wireless Networks with an Efficient Path Pruning Algorithm Presented by Min-Te Sun Department of Comp. Sci. & Software Eng. Auburn University

2. Outline • Introduction to geographical routing • Motivation of our research • Path pruning algorithm • Example • Simulation results • Conclusion and future work

3. Routing in Wireless Networks • Multi-hop wireless networks • Each node has limited wireless transmission range • A packet may travel through multiple intermediate nodes before it reaches its destination • Routing protocol governs where to forward the packet • Ideal routing protocol for wireless networks should be light-weight i.e., requires less amount of resources • Low communication overhead • Low storage requirement • Low delay

4. Different Routing Protocols for Wireless Networks • Proactive (e.g., DSDV) • Each node actively maintains a routing table • Reactive (e.g., DSR, AODV) • A query is broadcast by the source for a route to the destination whenever a packet is generated • Hybrid (e.g., ZRP) • Proactive routing for nearby destinations and reactive routing for others • Geographical

5. Geographical Routing Protocols • Assumptions • Each node knows the geographical locations of itself and its immediate neighbors • When a packet is generated, the source knows the geographical location of the destination • No global state is required • Scalable

6. Skeleton of Geographical Routing Protocols • Greedy forwarding - If a node holding a packet has a “better neighbor” to forward the packet, forward the packet to the best one. • Detouring strategy – If a local minimum is reached, a “backup plan” is required to find a detour that leads the packet away from the local minimum and toward the destination. • Depending on the definition of “better neighbor” and “backup plan”, we have different geographical routing protocols.

7. Typical Definition of “Better Neighbor” • The neighbor closest to the destination (GPSR) • The farthest node within a predefined angle span (compass routing) • The neighbor with the max value of PRR×DIST (Energy-Efficient Forwarding Strategy) • All the above better neighbor definition will lead to suboptimal route if the destination can be found by using only the greedy forwarding

8. Backup Plans for Finding Detour • Flooding (LAR) • Face routing (AFR, OFR, GOAFR+) • Perimeter routing (GPSR) • Except flooding, most of these approaches require the topology to be planarized to avoid possible infinite loop • Some edges need to be removed from routing consideration to make sure the topology contains no cross edges • GG, RNG, Planar Spanner, Morelia test

9. Impact of Planarization • Most of these distributed planarization algorithm favor short edges over long ones • Removing edges can potentially • disconnect the network • Luckily not likely to happen in practice • remove “shortcuts” between source and destination • A necessary evil!

10. Thoughts on Geographical Routing • Greedy forwarding is good enough if it alone can produce a path • If detouring mode (backup plan) is involved in routing, then • flooding is too expensive • non-flooding strategies (e.g., perimeter routing and face routing) tend to produce too many hops due to simple loops or not taking obvious shortcuts

11. Example Network Topology

12. Topology after Planarization (RNG)

13. Routing Path via GPSR

14. Our Goals • If no backup plan is involved when searching for route to the destination, then use the route found by greedy forwarding • It has been shown to be sub-optimal • If only one packet is sent, then send it with the route found in the planarized network topology • If more than one packet will be sent, we should look for shortcuts for the subsequent packets • Two types of shortcuts • those deleted by planarization (e.g., <A, C>) • those overlooked due to the heuristic traversal method after planarization (e.g., <D, G>)

15. Geographical Routing w/ Path Pruning • Greedily find the “best neighbor” recursively • If destination is reached, then use the route found by greedy mode • If not, construct the planar graph and find the best route as the original routing protocol for the first packet • After forwarding a packet, each node passively listens and see if any neighbor other than the one it previously forwarded the packet to is sending the same packet. If such neighbor is found, set the next hop to that neighbor • The following packets will then be forwarded directly to the recorded next hop

16. Topology after Planarization (RNG)

17. Routing Path via GPSR

18. Illustration of Path Pruning

19. What Exactly Are We Doing? • In our pseudo code, the planar graph is used to search for a detour to the destination. However, once a detour is found, our strategy will identify shortcuts immediately after the first packet is sent. • In other words, our strategy allows each node to utilize: • the edges deleted in the planarization • the edges not considered in the detouring plan (e.g., right-hand rule) • These edges are now acting as shortcuts!

20. What Price Do We Pay? • No extra power consumption • Each node has to passively listen to the channel for possible packet for it anyway • No communication overhead • No extra packet is sent • Only local location information is used • Slight storage overhead • At most one entry per node per active connection • Not every node on the route has to remember the next hop

21. Properties of Path Pruning Algorithm • The algorithm does not change the routing path if only greedy forwarding is involved. • The node set of the pruned route is a subset of the one from the original route. • In a pruned path, if two nodes are in the same mode but not consecutive on the route, then they can not be neighbors. • The path pruning algorithm converges when the first packet reaches the destination. • A pruned path is loop free.

22. C-shape wireless networksGPSR vs GPSR w/ Path Pruning GPSR: 68 hops GPSR with PP: 11 hops Effective when networktopology contains largevoid area

23. C-shape wireless networksGOAFR+ vs GOAFR+ w/ PP GOAFR+: 47 hops GOAFR+ w/ PP: 10 hops

24. Simulation Settings • Four protocols: GPSR, GOAFR+, GPSR w/ Path Pruning, GOAFR+ w/ Path Pruning • Transmission range is 1 unit • Nodes are placed randomly with uniform distribution in a 20 by 20 unit region. • Number of nodes ranges from 100 to 1900 • Corresponds to 0.79 to 14.9 nodes per unit disk of area π • 2000 realizations of network topologies with random source and destination are generated for a given network density • Assume no collision (separate impact of MAC)

25. Measurement of Routing Algorithm Performance • Performace • Average Performance

26. Average performance of routing algorithms (GG planarization)

27. Average performance of routing algorithms (RNG planarization)

28. Average temporary and steady-state overhead of path pruning (GG planarization)

29. Conclusion • Path pruning can efficiently fix the lengthy detour route issue in non-flooding based geographical routing protocols with the help of few state info passively maintained at a subset of nodes on the path. • Our protocol allows the following packets to traverse farther using the same number of hops • If a packet is dropped due to TTL expiration, source does not need to set a larger TTL value for retransmitted packets to reach the destination

30. Delivery Rate vs TTL and # of Transmissions (550 nodes, GPSR)

31. Questions?