Multicast in Wireless Mesh Network

1. Multicast in Wireless Mesh Network Xuan (William) Zhang Xun Shi

2. Outline • Introduction to multicast in WMNs • Defining the cost of multicast tree • Ruiz’s MNT protocol • Chou’s MDM protocol • Conclusion

3. Outline • Introduction to multicast in WMNs • Defining the cost of multicast tree • Ruiz’s MNT protocol • Chou’s MDM protocol • Conclusion

4. What is Multicast? • “Point-to-multipoint" or "multipoint-to-multipoint“ • Different from broadcast and unicast (a) Broadcast (b) Multicast (c) Unicast

5. Advantages of Multicast • Delivery to destinations simultaneously • Deliver the messages over each link of the network only once • Only create copies when the links to the destinations split

6. Wireless Mesh Networks • Mesh routers are generally stationary • Multi-hop forwarding • High speed • Reliable power supply

7. Internet multicast protocols • Feature • Wired / Powerful / Reliable • Maintain a large and fixed topology • Shortest path algorithms • simpler to implement • simpler to support frequent joins/leaves • lowest delay

8. Drawbacks of Internet multicast in WMNs • Routing metrics do not aim at minimizing the cost of multicast tree • Not using broadcast nature

9. MANET multicast protocols • Feature • Maintaining a smaller and mobility network topology • Relying on flooding mechanism • On-demand routing protocols • Suitable for mobility • Low power consumption

10. Drawbacks of MANET multicast in WMNs • Complexity of computation • High mobility • High Power consumption

11. Multicast protocols in WMNs • WMNs multicast is between Internet and MANET multicast • Fixed topology • Broadcast nature • Mobility and power are not problems

12. Outline • Introduction to multicast in WMNs • Defining the cost of multicast tree • Ruiz’s MNT protocol • Chou’s MDM protocol • Conclusion

13. Traditional definition of cost • Measured by hops, delays, etc. • Minimum Steiner tree problem • NP-complete • Heuristic algorithms – polynomial time • Shortest path tree • Sub-optimal shared tree • MST algorithm: 2*optimal approximation • Zelikovsky algorithm: 11/6*optimal approximation

14. Define the cost in WMNs • Cost: number of transmissions • Minimize the number of transmissions • Maximize the forwarding nodes which are shared by sender-receiver paths • This problem is NP-complete

15. Problem with Steiner Tree • Steiner Tree: minimum edge cost • Broadcast: node can send neighbors data in one transmission • Our goal: minimizing the number of transmissions!!

16. Outline • Introduction to multicast in WMNs • Defining the cost of multicast tree • Ruiz’s MNT protocol • Chou’s MDM protocol • Conclusion

17. Ruiz’s Algorithm • Purpose: find minimal data overhead tree • Contributions: • Theorem 1: Prove Steiner tree is not optimal in WMNs with respect to the number of transmissions • Theorem 2: Prove minimal data overhead tree is NP-Complete • Proposed heuristics to compute trees with minimizing the number of transmissions

18. Problem statement • Define t is multicast delivery tree • Define Ct(t) is the number of transmissions required to deliver a message from sender s to receiver set R • Problem statement: Minimize the Ct(t) • Ct(t)=1+|Ft| • Minimize the number of forwarding nodes

19. Theorem 1: Steiner tree not minimal • Steiner multicast tree (minimal edge cost) is not the minimal data-overhead multicast tree. • Proof by example:

20. Theorem 2: NP-Complete • Proof by including a particular case • Special case: R=V-{s}, find the smallest forwarding nodescovers the rest of nodes in V-{s} Vertex cover problem – NP-complete

21. Heuristic Algorithm • Goal: approximate minimal data overhead multicast tree • Reduce the number of forwarding nodes • While increase the number of leaf nodes • Centralized greedy-based heuristic algorithm • Distributed heuristic algorithm

22. Greedy minimal data overhead Alg. • Centralized WMNs • Greedily build cost-effective sub-trees • A node v is selected a forwarding node only if it covers two or more nodes

23. Greedy minimal data overhead Alg. cont. • Steps • Construct a cost-efficient sub-trees • Build a Steiner tree among the roots of the sub-trees

24. Initialize Loop V=V-{s} V=V-v aux=aux-Cov(v)+{v} aux=R-Con(s)+{s} MF=MF+{v} empty R6 R6 R6 R6 R6 R1 R1 R1 R1 R1 R2 R2 R2 R2 R2 M1 M3 M1 M2 M1 M3 M2 M3 M1 M2 M1 M3 M3 M2 M2 S S S S S R4 R4 R4 R4 R4 R5 R5 R5 R5 R5 R3 R3 R3 R3 R3 Alg Demo Stop!! All nodes in V now only cover at most 1 receiver V (unvisited nodes) M3, M1, R1, R2,R3, R4, R5, R6 M3 M2 M1, R1, R2,R3, R4, R5, R6 M2, M3, M1, R1, R2,R3, R4, R5, R6 aux (nodes to cover list) minimal data overhead tree! Hehe!! S, R5, R6, M2 S, M2, M3 S, R2, R3, R4, R5, R6, MF (multicast forward node list) M2 M2, M3 ε MST heuristics to build Steiner tree

25. Performance Evaluation • Compared Algs • SPT: source path tree Alg • MST: Steiner tree Alg • MNT: centralized proposed Alg • MNT2: distributed proposed Alg • Simulations • Number of Tx required • Mean number of hops • Number of Tx with density

26. Performance Evaluation cont. • Number of transmissions required The total number of packets transmitted either by the source or any relay node in path. MNT, MNT2 MST SPT Theorem 2, Steiner tree is not minimum data-overhead. Do not aim at minimize the cost of the tree.

27. Performance Evaluation cont. • Mean path length (Mean number of hops) The number of multicast hops from a receiver to the source averaged over the total number of receivers. MNT, MNT2 MST SPT Aim at minimize the length of the tree.

28. Performance Evaluation cont. • Number of transmissions with density Examine reduction of Tx numbers when increase the density. Proposed heuristic MNT, MNT2 reduced more than SPT and MST!

29. Summary of Ruiz’s Algorithm • Steiner tree does not suitable in WMNs • The proposed Algorithm is NP-complete • Heuristic Algorithm • Centralized Algorithm • Distributed Algorithm • Evaluation • the higher the density, the higher are the Heuristic Alg performance

30. Outline • Introduction to multicast in WMNs • Defining the cost of multicast tree • Ruiz’s MNT protocol • Chou’s MDM protocol • Conclusion

31. Resilient Forwarding Mesh • Makes multicast robust to node or link failure • 2 paths • Increases PDR and throughput

32. Resilient Forwarding Mesh Example (a) Network topology (b) Optimal solution (c) Suboptimal solution

33. Node-Disjoint Paths • Parallel routes that connect the source and the destination • Do not have any node in common except the source and destination • Deliver packets simultaneously

34. Optimal Resilient Forwarding Mesh • Each source-destination pair is connected by two node-disjoint paths • Total number of broadcast transmissions is minimized • Minimizing the number of broadcast transmissions is NP-complete • Use heuristic algorithms to obtain approximate solutions

35. Heuristic Approximation Algorithms • Tree-based • Node-Disjoint Tree Algorithm (NDT) • Revised Node-Disjoint Tree Algorithm (RNDT) • Path-based • Shared Disjoint Mesh Algorithm (SDM) • Minimal Disjoint Mesh Algorithm (MDM)

36. Node-Disjoint Tree Algorithm (NDT) • Build a multicast tree PT with minimal number of transmissions using the MNT • Remove all intermediate nodes of PT from node set V • Find a new minimal multicast tree BT in the new V • Add all intermediate nodes of PT and BT to RFM

37. NDT Example S S S M1 M1 M2 M2 M3 M3 M3 R1 R1 R1 R2 R2 R2

38. NDT Example S S S M1 M2 M1 M2 M3 M3 M3 R1 R2 R1 R2 R2

39. Shared Disjoint Mesh Algorithm • Find a shortest path P • Remove all intermediate nodes of P from V, and find another shortest path B which is node-disjoint to P • Update out-flow links of all intermediate nodes to zero • Add all intermediate nodes of PT and BT to RFM • Repeat above steps for all receivers

40. SDM Example S 1 2 2 0 0 M1 M2 M2’ 2 2 0 0 M3 1 2 2 0 0 2 2 0 0 R1 R2 5 5

41. Minimal Disjoint Mesh Algorithm • Improves SDM in the way of building the node-disjoint path pair • Use Suurballe’s algorithm to find node-disjoint path pair with minimal cost at the same time

42. Suurballe’s Algorithm Example S S 1 10 1 10 1 1 M1 M2 M1 M3 M2 M3 1 10 1 10 10 1 10 1 100 100 R R Cost = 3 + 101 Cost = 11 + 12

43. Comparison of the 4 Protocols • Simulated in QualNet • Manually calculate optimal solution up to session size of 10 • Performance is measured by the number of transmissions as a function of multicast session size

44. Performance Comparison NDT Number of Transmissions RNDT SDM MDM Multicast Session Size

45. Summary • NDT and RNDT are tree-based heuristic algorithms • SDM and MDM are mesh-based heuristic algorithms • MDM used Suurballe’s algorithm to find node-disjoint path pair with minimal cost • Total Number of transmissions: MDM<SDM<RNDT<NDT

46. Compare MNT with MDM

47. Compare MNT with MDM cont. • MDM needs additional transmissions to provide resilience • MDM needs more transmissions when session size is small • When session size increases, the MDM is more likely to find the disjoint paths that share more common intermediate nodes

48. Outline • Introduction to multicast in WMNs • Defining the cost of multicast tree • Ruiz’s MNT protocol • Chou’s MDM protocol • Conclusion

49. Lecture Summary • Ruiz’s • The MNT is NP-complete • Heuristic Algorithm • Centralized Algorithm • Distributed Algorithm • Chou’s • Tree-based: NDT and RNDT • Path-based: SDM and MDM • Total number of transmissions: MDM<SDM<RNDT<NDT

50. References • Heuristic algorithms for minimum bandwidth consumption multicast routing in wireless mesh networks, P. M. Ruiz, and A. F. Gomez-Skarmeta, Proceedings of ADHOC-NOW, 2005. • Protecting Multicast Sessions in Wireless Mesh Networks, X. Zhou, J. Guo, C.T. Chou, and S. Jha, IEEE Conference on Local Computer Networks, 2006. • Simulation Study of Diverse Routing and Protection Algorithm in Mesh WDM Network, X. Yao, and C. Chen, 2004. • A Performance Comparison Study of Ad Hoc Wireless Multicast Protocols, S.J. Lee, W. Su, J. Hsu, M. Gerla, and R. Bagrodia, Proceedings of IEEE INFOCOM, 2000. • A Fast Algorithm for Steiner Trees, L. Kou, G. Markowsky, and L. Berman, Acta Informatica, No. 15, vol. 2, 1981, pp.141-145.