1. Reducing Broadcast Latency in Wireless Mesh Networks (WMNs) Cyrus Minwalla Maan Musleh COSC 6590

2. 2 Presentation Layout Overview Broadcasting in wireless mesh networks (WMNs) Broadcast configurations in WMNs: Fully multi-rate multicast (FMM) Single “best-rate” multicast (SBM) Performance Evaluation Conclusion

3. 3 Brief overview of Wireless Mesh Networks (WMNs) Network Topology Properties of WMNs

4. 4 Network Topology in WMNs

5. 5 Properties of Wireless Mesh Networks Nodes: Wireless but static Connected in an ad-hoc manner Energy a non-issue (nodes generally plugged in, or easily rechargeable) Network: Topology is cluster-based: Static routers connect subsets of the network. Routers can serve as source nodes for sub- trees (useful for topology construction, scheduling, etc.)

6. 6 Why Broadcasting in WMNs Motivation: Carried over from wired networks Useful for many applications: OS updates Video conferencing/streaming Multiplayer gaming Have fewer packet transmissions due to “wireless broadcast advantage” (WBA)

7. 7 What is Wireless Broadcast Advantage (WBA)? Refers to a unique quality belonging to wireless networks Wired networks perform broadcast by separate unicasts across the network (separate to each root node in a tree) In wireless networks: Direct neighbours of the source node require only one tx Multiple unicast tx in wired = 1 broadcast tx in wireless Potential Energy and bandwidth savings!

8. 8 Exploiting WBA for Broadcast Achievement of WBA in broadcast transmissions  configuration changes at link level Link level changes involve: Number of radios/channels Rates Radio power (for channel reuse) Antenna gain (direction)

9. 9 Node Configuration Various node configurations in literature Authors discuss the following two configurations: Single-radio single-channel multi-rate Multi-radio multi-channel multi-rate

10. 10 What is Minimum Latency Broadcasting (MLB)? Definition: To provide the best QoS by minimizing latency at the slowest node Goal: All destination nodes must receive packet within same time frame Maximize the transmission rate of the slowest node Metric: RAP (Rate-Area Product)

11. 11 Why do we care about MLB? Motivation: Want to guarantee quality of service (QoS) to all users in the multicast session Want to decrease the latency encountered by the slowest link.

12. 12 Overview of Techniques Both techniques involve the idea of using multicasts across partitioned nodes to achieve broadcast Single-channel multi-rate: Also known as “fully multi-rate multicast” (FFM) Multi-channel multi-rate: Referred to as the “single best-rate multicast” (SBM)

13. 13 Multi-rate vs. Multi-radio FMM: Uses an optimum rate per link to maximize throughput and minimize latency Attempts to minimize the number of transmissions Needs scheduling per transmission to avoid interference SBM: Determines a single best-rate metric for the entire network Simplifies the construction algorithm by using one rate Uses multiple channels, thus simplifying the scheduling algorithm

14. 14 What about Energy Efficiency? Both techniques transmit a packet multiple times from the same node: Multi-rate uses multiple rates for various neighbours (based on RAP) Multi-channel uses multiple channels (channel diversity  non-interference) The goal: To minimize broadcast latency, not energy efficiency

15. 15 Fully Multi-rate Multicasting (FMM)

16. 16 Fully Multi-rate Multicasting (FMM) Topic Layout: What is fully multi-rate multicast? Why we want to use it How it works Topology Construction Algorithm Multicast Grouping Algorithm (Simplified) Transmission Scheduling Maximum end-to-end throughput Pros and Cons Recap

17. 17 What is “Fully Multi-Rate Multicast” ? Broadcast achieved via sequential multicasts Multicast to separate subsets in network Algorithm in four steps: Construct a tree to span the entire network Calculate the optimum rate at every link Provide scheduling for all transmissions Recalculate maximum end-to-end throughput Caveat: Most of the solutions are NP-hard

18. 18 Why choose FMM Motivation: Multi-rate allows us to minimize the MLB Current radios work with setup RAP metric is easy to calculate

19. 19 Current 802.11 metrics Transmit rates and ranges for 802.11b Obtained via Qualnet simulation Consider network topology in next slide

20. 20 A Motivational Example Node 1 wants to broadcast to 2, 3, 4 and 5. Node 2 @ 11 Mbps, node 5 @ 1 Mbps One single transmission at lowest rate or two transmissions (one at either rate)

21. 21 Motivational Example: The Single Transmission Case Node 1 broadcasts to nodes 2 and 5 Transmission rate = slowest link i.e. 1Mbps Transmission to node 2 @ 1Mbps  4 is starved until 33 u.t.

22. 22 Motivational Example: The Multiple Transmission Case Node 1 makes two transmissions Transmission 1 to node 2 @ 11 Mbps Transmission sequence: 2  3  4 Node 1  5 only occurs when 2  3 is complete Node 4 receives packet at 23 u.t.

23. 23 Topology Construction in FMM We want to reach all nodes within the network: Build a connected dominating set (CDS): Def’n of CDS: In a graph G(v,e), the connected-dominating set is a set of edges S{e} | all non-leaf nodes v are connected. All other (leaf) nodes are one hop away from at least one node in CDS

24. 24 Connected-Dominating Set (CDS) What this means: In a CDS, the source has a path to all relaying nodes in the network Calculate all possible CDSs in the network Obtain the CDS with the minimum cost Steps: Calculate the set of possible CDSs Attach a cost metric per CDS Pick one that minimizes that cost (use Djikstra)

25. 25 Problems with CDS Problem 1: For k nodes, 2 k possible sets to consider Solution: Use Djikstra with an approximation criteria Problem becomes polynomial Problem 2: Minimum connected set will assume slowest rate to maximize downlink neighbours per node Same as using slowest rate for all nodes Solution: Account for the rate metric: max (no. of nodes x transmission rate) This is defined by the RAP

26. 26 Topology Construction in FMM Algorithm steps: Keep a set C of all covered nodes. C starts with just source node s Pick optimal product of rate x no. of nodes covered Add covered nodes within optimal area to C Continue until C satisfies CDS quality for G This process ensures a minimum-cost, minimum-spanning tree

27. 27 Sample Network Topology

28. 28 Example: Minimum WCDS Tree

29. 29 Example: Minimum WCDS Tree with rates

30. 30 Multicast Grouping in FMM Once the broadcast tree is constructed, need to determine two things for each node: No. of times to multicast No. of nodes covered by multi-cast Need to find transmission delay to reach all downstream nodes with minimum latency Every node’s latency depends on what happens downstream  follow bottom-up topology

31. 31 Bottom-up Topology Algorithm Steps: Start with leaf nodes Calculate the minimum latency to the relay (based on optimal rate in previous step) Latency maximum at relay node is stored in Cardinality Value (CV) CV helps determine the transmission delay at relay node R

32. 32 Example: Multicast Grouping

33. 33 Example: Multicast Grouping

34. 34 Example: Multicast Grouping

35. 35 Example: Multicast Grouping

36. 36 Example: Multicast Grouping

37. 37 Bottom-up Topology (2) CV values along nodes build up a transmission sequence For k rates, there are 2 k-1 possible valid transmission sequences (VTS) Pick the VTS with the shortest possible transmission delay Assumption Grouping does not deal with nodal interference

38. 38 (Simplified) Transmission Scheduling Transmission sequence determined by CV Higher CV = higher latency  more critical transmission All nodes assigned a start-time and a stop time Nodes must have packet before start time The goal is to avoid nodal interference In our example, time is measured in packet time: Packet tx @ 11 Mbps = 1 u.t.

39. 39 Example: Transmission Scheduling

40. 40 Example: Transmission Scheduling

41. 41 Example: Transmission Scheduling

42. 42 Example: Transmission Scheduling

43. 43 Example: Transmission Scheduling

44. 44 Problems with Transmission Scheduling Problem 1: Absolute times require centralized clock Solution: Algorithm assumes a centralized clock within source node Problem 2: Node schedules are broadcast throughout the network.. to set up broadcasting Solution 2:...........

45. 45 Pros and Cons Advantages Obtains lower latency compared to standard techniques Works with current hardware Disadvantages: Algorithms are NP-hard Scheduling problem has no apparent solution

46. 46 Recap The technique FMM: uses selected multicasts to achieve broadcast over network Minimizes latency in the network Algorithms required to achieve optimal solution = NP-hard Need a centralized station for clock synchronization + scheduling The next technique resolves some of these issues

47. 47 Single Best-Rate Multicasting (SBM)

48. 48 Single Best-rate Multicast (SBM) Decides a single transmission rate for all link layer data multicast. Depends on the network's topological properties. Simplifies broadcasting algorithms.

49. 49 Decisions To Be Made Selecting 'best' transmission rate to use for all link layer broadcasts. Deciding whether a certain node should transmit. Deciding 'Interface Grouping'. Scheduling each node's transmissions.

50. 50 'Best' link-layer multicast rate selection Can be predicted reasonably by the product of the transmission rate and transmission coverage area (rate-area product or RAP). Higher RAP means more broadcast- efficient for SR-SC MR WMNs.

51. 51 Methods of Selection  R => set of transmission rates, which if used returns a connected network. 1.Use the highest link-layer multicast rate in R. “Quickest rate”. 2.Use the transmission rate with the highest RAP value of all Rates in R.

52. 52 Topology Construction Two Heuristics Proposed 1.Connected Dominating Set (CDS):  Simplified Minimum Connecting Dominating Set Problem. 2.Parallelized Connected Dominating Set (PCDS)  Adaptive to the radio resources available (interfaces and channels).  Uses two more parameters ( priority and label).

53. 53 Example – CDS Construction

54. 54 Interface Grouping and Transmission Scheduling Broadcast performance can be improved by delaying the choice of interface to use till the scheduling stage WMN can then maximally exploit the channel diversity in the system.

55. 55 Interface Grouping and Transmission Scheduling During scheduling, an appropriate choice of the interface to use is made Depending on other transmissions at that time The algorithm aims to find a start time and end time of each transmission node For this algorithm, nodes are sorted in descending order according to height of node. Height is distance from the node to its furthest leaf.

56. 56 Interface Grouping and Transmission Scheduling The choice of channel to be used for a particular transmission is motivated by the desire to include as many parallel transmissions as possible. The algorithm completes execution when all transmissions are scheduled.

57. 57 Normalized Broadcast Latency

58. 58 Review of Presentation Topics Covered: Broadcasting in WMNs What is WBA? What is MLB? Techniques with examples: Fully multi-rate multicast (FMM) Single best-rate multicast (SBM) Performance Comparison

59. 59 Future Work Sleep… Actually, to find a feasible solution for the scheduling algorithm

60. 60 Bibliography [1] C.T.Chou, A. Misra and J. Qadir. Low latency broadcast in multi-rate wireless mesh networks. IEEE JSAC special issue on wireless mesh networks, 2006. [2] J. Qadir, C.T.Chou and A. Misra. Exploiting rate diversity for multicasting in multi-radio wireless mesh networks. IEEE, 2006. [3] R. Draves, J. Padhye, and B. Zill. Routing in multi-radio, multi-hop wireless mesh networks. In Mobicom, pages 114- 118, 2004 [4] H. Lim and C. Kim. Flooding in wireless ad hoc networks. Computer Communications, 24(3-4): 353, 2001

61. 61 Thank you for your time and patience Questions/Comments?