LANMAR: Landmark Routing for Large Scale Wireless Ad Hoc Networks with Group Mobility

1. LANMAR: Landmark Routing for Large Scale Wireless Ad Hoc Networks with Group Mobility Guangyu Pei, Mario Gerla and Xiaoyan Hong

2. Outline • Introduction • Ad hoc routing • Link State Routing • Distance Vector Routing • Fisheye State Routing • LANMAR • Routing • Drifting and isolated nodes • Simulations • Related works • Landmark election • Dynamic group discovery • M-LANMAR • Direction forwarding

3. Ad-Hoc routing protocol • A wireless network between mobile nodes • No infrastructure • Limited communication distance • Usually low-powered • Mult-hop communication • How to route messages? • Flooding • Simple – keep track of neighbors only • Inefficient use of bandwidth • Not scalable • Along the (shortest) path from source to destination • Requires nodes to keep some topology data

4. Mobility • Node moves cause network topology changes • Results in routes change • Nodes need to update their topology information • A node (dis)connecting to(from) a neighbor needs to notify other nodes • How to make update process efficient • Low overhead • Fast update propagation

5. Routing algorithms • Proactive • Constantly monitor network topology • Advantage: Always know how to route to any node (up to mobility) • Disadvantages • Routing table storage • Periodic control traffic overhead (to handle mobility) • Example protocols: LSR, OLSR, DSDV, FSR, LANMAR • Reactive / on-demand • Look for a route only when need to send a packet • Advantages • Low routing table storage (only routes in use) • No periodic control traffic • Disadvantage: route search overhead with high mobility and many short lived flows • Example protocols: AODV, TORA, DSR, ABR

6. Link State Routing • Similar to existing protocols of Internet routers • Routing table update at each node • Periodically or on link addition/removal: flood “link state” – list of neighbors • Re-broadcasts link state information received from neighbors • Employ “sequence numbers” to distinguish new from stale updates • Each node maintains a routing table with all the “link state” of all the nodes in the system • Routing • The destination is stored in the message header • Each forwarding node finds the shortest path to the destination according to its routing table

7. 1 Link State Routing 0 {1} • At node 5, based on the link state packets, topology table is constructed: • Dijkstra’s algorithm can then be used for the shortest path {0,2,3} {1,4} 3 2 {1,4,5} 4 {2,3,5} 5 {2,4}

8. Link State Routing drawbacks • As number of nodes grows / mobility increases • Routing tables grow linearly (assuming constant density) • Link state control traffic grow linearly • Not scalable

9. Distance Vector (DV) Routing • Do not keep the complete links state of node i • Only the next hop neighbor and the distance towards i • Routing is simpler (no need to find a shortest path) • Maintenance • Periodically exchange distance vector (DV) with neighbors • Update own DV if receive a shorter path to i • Problems • Loops • Count to infinity • Scalability – similar to LS

10. Fisheye State Routing • A variant of Link State routing • Aimed at reducing control traffic (link state updates) • At the expense of routing table accuracy • Which entries in the routing table can be updated less frequently? • The ones corresponding to a more distant nodes

11. Fisheye State Routing • Maintain accurate information in immediate neighborhood (hop=1) • Progressively less detail as distance increases • Entries of nearby nodes are exchanged more frequently • Frequency decreases proportionally to distance • Topology information is exchanged between neighbors via Unicast • Routing - as packet gets closer to destination, routing accuracy increases

12. Fisheye State Routing LST HOP 0 LST HOP 0:{1} 1:{0,2,3} 2:{5,1,4} 3:{1,4} 4:{5,2,3} 5:{2,4} 1 0 1 1 2 2 0:{1} 1:{0,2,3} 2:{5,1,4} 3:{1,4} 4:{5,2,3} 5:{2,4} 2 1 2 0 1 2 1 3 Entries in black are exchanged more frequently LST HOP 2 0:{1} 1:{0,2,3} 2:{5,1,4} 3:{1,4} 4:{5,2,3} 5:{2,4} 2 2 1 1 0 1 4 5

13. FSR – Conclusions • Major scalability benefit: control traffic decreases significantly • Unsolved problems • Route table size still grows linearly with network size • Out of date routes to remote destinations

14. LANMAR

15. Logical Group Landmark LANMAR • Define an ad hoc groups of nodes moving together • Fisheye routing inside the group • Other routing algorithms possible • Landmark node elected for each group • Each node maintains • FSR routing table inside its group • Accurate route to all landmarks • Association of each node to its group (landmark)

16. Routing • Definitions • Logical subnet: a group of nodes moving together • Node logical address: <subnet, host> • Node local scope: a k-hop neighborhood of the node • Routing based on existing routing protocols • FSR routing within local scope • Distance Vector routing to distant nodes via landmarks • Reduction of both control overhead and table size • Scalable to larger networks

17. Routing tables • A – neighbor list • TT – Fisheye routing table for local scope • j  {all the nodes in the scope} • LS(j) – Link state of node j • SEQ(j) – Link state timestamp of node j • NEXT – next hop table • j  {all the nodes in the scope}  {all the landmarks} • NEXT(j) – the next hop along shortest path • D – distance table • j  {all the nodes in the scope}  {all the landmarks} • D(j) – shortest path length to j • Other distance functions possible (e.g., bandwidth)

18. Routing algorithm • If the destination in the scope • route by TT • Else • Set landmark = a landmark node in the subnet of the destination • Route towards landmark by NEXT • Packets do not need to pass through the landmark • Once reached the scope of the destination, routed directly via Fisheye tables (TT) • Do not overload landmark

19. Updating routing tables • Similar to FSR - periodically exchange scope topology with neighbors • Assume all the subnet is within the scope of its landmark • Piggy-back landmark distance vector • No details in the paper on NEXT and D maintenance

20. Landmarks • Landmark election not described in the paper • Assume some additional algorithm • No support for moving between groups/landmarks • Would require changing subnet – effectively a new node

21. Routing table storage • 100 nodes, 4 groups • FSR: 2600 bytes per node • LANMAR: 690 bytes per node • N nodes, N subnets, N nodes in scope • FSR: O(N) per node • LANMAR: O(N) • Also decreases control traffic, power consumption

22. Drifter nodes • Assume most of the subnet is within the scope of its landmark • Few nodes may move out of the scope • The landmark L needs to know a path to a “drifter” k • Modify the routing protocol • Each node i, on the shortest path between L and k, keeps a DV entry to k • If k is within the scope of i, it is already in i’s Fisheye table • When i transmit its DV to j (in the same subnet), j keeps k’s entry iff • d(j,l) < scope OR d(j,L) < d(i,L) • Landmark has a path to all drifters • For 20% of drifter nodes ~7% extra overhead (in some specific setting)

23. Isolated nodes • Nodes in groups of size 1 • If such nodes are rare – consider them landmarks • If many isolated nodes – need different protocols • Hybrid protocol • Consider isolated nodes landmarks • Lower DV update frequency with distance increase • Gradual transition from LANMAR to FSR performance • On-demand routing to isolated nodes • Associate each isolated node with Home Agent • Home Agent constantly maintains a route to a node

24. Illustration

25. Simulations • Setup • 1000 x 1000 meters square • 150 meter range • 2 Mbit/sec channel capacity • 100 nodes, 4 groups • Scope = 2 hops • Source-destination pairs chosen randomly • UDP messages – 512 bytes, sent every 2.5 seconds • Reference Point Group Mobility • Two components – individual and group • Each based on a random waypoint model • Speed - 2 to 10 m/s

26. Performance metrics • Routing effectiveness metrics • Packet delivery fraction • Average end-to-end packet delay • Not independent – dropped packets not included in delay computation • Normalized routing load • Number of control packet per delivered data packet • Throughput – the actual throughput at destination

27. Delivery fraction • Only 10 pairs communicate • On-demand protocol(AODV) performs best • LANMAR outperforms FSR as speed increases • Better to have accurate routes to few landmarks that inaccurate ones to all distant nodes

28. Delivery fraction • 300 pairs communicate • On-demand protocols underperform due to control traffic lost to buffer overflows • FSR and LANMAR nearly unchanged • LANMAR outperforms all other protocols beyond 30 pairs

29. Delay • As load increases delay increases due to queue buildup • LANMAR performs best due to low control overhead

30. Routing overhead • For FSR and LANMAR the overhead is constant • Normalized load (overhead per delivered packet) rises as delivery fraction falls • On-demand protocols’ load increases sharply with traffic

31. Throughput • Throughput grows with load until the network saturates • Depends also on delivery fraction • AODV saturates first, due to high routing overhead • LANMAR outperforms other protocols

32. Conclusions • LANMAR improvements over FSR • Lower storage and traffic overheads • Better performance under high load/mobility • Assumption: nodes move in groups • Instead of handling each distant node individually, handle as a group • Need additional algorithm for landmark election • Yet another hierarchical protocol (ZRP, HSR, CGSR) • Not really scalable – one level of hierarchy • Non-adaptive groups/FSR scopes • √N group/scope size analysis does not hold for dynamic joins/leaves

33. Related Work

34. Landmark electiondraft-ietf-manet-lanmar-03.txt • Landmark election algorithm: • No landmark exists initially, only FSR progresses • A node proclaims itself as a landmark when it detects more than T group members in its FSR scope • An election is required to select the winner in the group • Election algorithm • A node with the largest number of group members wins and the lowest ID breaks a tie • To prevent frequent landmark change • The current election winner replaces the old landmark when its number of group members is larger than the old one by an extra fraction • Or, the old landmark gives up the landmark role when its number of group members reduces to a value smaller than a threshold T

35. Dynamic group discovery"Dynamic Group Discovery and Routing in Ad Hoc Networks", X. Hong and M. Gerla • Groups are not known in advance • Location and speed not required (no GPS) • Each node finds its Traveling Companions (TCs) • By inspecting FSR tables for some time window W • Leader (landmark) election • Node’s “weight” is the number of its TCs • Each node broadcasts it weight to its group • Node with highest weight is elected • Leader’s ID become the subnet ID • Works only initially, group changes require more work

36. M-LANMAR “Scalable team multicast in wireless ad hoc networks exploiting coordinated motion”, Y. Yi, M. Gerla, K. Obraczka

37. M-LANMAR • Multicast: transmit the same packet to multiple destinations • Unicast to each destination is inefficient - wastes network bandwidth • Intra – group multicast: • Sensor data “fusing” • Inter – group multicasting: • “fused” video/image/data is multicast to other groups

38. M-LANMAR • Based on LANMAR • Nodes move in groups • Landmark is elected in each group • Multicast LANMAR (M-LANMAR) • Unicast tunneling from the source to the Landmark of each “subscribed” group • Scoped flooding within a group

39. Scope = 2 Flooding Tunneling Scope = 2 Flooding LM2 LM1 LM3 Subscribed Teams Source node LM4

40. DFR “Direction” Forward Routing for Highly Mobile Ad Hoc Networks, YZ Lee, M Gerla, J Chen, J Chen, B Zhou, A Caruso

41. DV • In DV routing, node keeps pointer to successor to destination Route update Successor Data flow Source Sink • When the successor moves, the path is broken • Alternate paths, even when available, are not used • Solution: direction forwarding

42. Direction Forwarding • Routing update creates not only “successor”, but also “direction” entry Route update Successor Data flow Source Sink “Direction” to Sink • Select “most productive” neighbor in forward direction • If the network is reasonably dense, the path is salvaged

43. How to compute the “direction”? • Need “stable” local orientation system (e.g., virtual compass) to determine direction of update • GPS will do, but it is an overkill (global orientation) • Several non-GPS local coordinate systems have been proposed • Sextant [Mobihoc ’05]; beacon DV • Local reference system must be refreshed fast enough to track average local motion

44. Computing the “direction” • Compute “direction” to a destination when the routing updates are received • The “direction” to the upstream neighbor is used as the “direction” to the destination • If multiple updates received from different neighbors with same hop distance • Take vector sum of directions

45. “Direction” to a destination C A B Computing the “direction” • Node A receives DV update packets from B & C • Compute the “directions” to node B & C, respectively, Directions to neighbors Computation of the “direction” • Unit vectors are used to combine the two “directions”

46. Delivery fraction increases DFR LANMAR Delivery ratio vs. speed (Excluding packet loss due to disconnected destination)

47. DFR Conclusions • DFR: new forwarding strategy for table driven routing • Direction Forwarding can improve traditional routing performance dramatically in high speed • DFR is about as robust as geo-routing • Yet DFR does not suffer the limitations of geo routing: • GLS • Global GPS required at all mobiles • Possible dead-end ineffciencies

48. Thank You