1. Opportunistic Flooding in Low-Duty- Cycle Wireless Sensor Networks with Unreliable Links Shuo Goo, Yu Gu, Bo Jiang and Tian He University of Minnesota, Twin Cities ACM MobiCom 2009

2. Outline  Introduction  Network Model and Assumptions  Main Design  Simulation  Conclusions

3. Introduction  Low Duty Cycle wireless sensor networks Long Network Lifetime t … Active State Dormant State

4. Introduction  Flooding in low-duty-cycle WSNs. No longer consists of a number of broadcasts. Instead, it consists a number of unicasts. Active State Dormant State B C D A B C D A t B CD

5. Introduction  Motivation Existing solutions are not suitable to be directly applied to low-duty-cycle wireless sensor networks. Unreliable wireless links

6. Introduction  Design Goal Fast data dissemination: shorter flooding delay Less transmission redundancy: less energy cost Three challenging issues C B A  Unreliable links  Redundant transmissions  Collisions

7. Network Model and Assumptions  Time synchronization of all sensor nodes.  Pre-determined working schedules shared with all neighbors.  Unreliable wireless links The probability of a successful transmission depends on the link quality q.  Flooding packets are only forwarded to a node with larger hop-count to avoid flooding loops.

8. Main Idea  Energy-Optimal Tree No redundant transmissions Long flooding delay

9. Main Idea  Adding opportunistically early links into the energy-optimal routing tree

10. Delay Distribution Computation 0.9 0.8 Probability mass function (pmf)

11. Decision Making Process 11111 Time B’s pmf 0.5 0.3 0.1 0.04 168 2432 Early PacketsLate Packets Early packets are forwarded to reduce delay Late packets are not forwarded to reduce energy cost For p = 0.8 Dp = 16 AB 0.5 p-quantile Dp A receives packet Expected Packet Delay (EPD)

12. Expected Packet Delay Computation 11111 Time B’s pmf 0.5 0.3 0.1 0.04 Dp= 16 168 2432 A is expected to transmit twice! A receives packet A’s first try to B A’s second try to B EPD = 24 EDP = 16x0.5 + 24x0.5x0.5 + 32x0.5x0.5 2 +… … AB 0.5

13. Decision Making Process 11111 Time B’s pmf 0.5 0.3 0.1 0.04 Dp = 16< EPD = 24. A will not start the transmission to B! 168 2432 Dp= 16 EPD = 24

14. p-quantile Dp Early Packets’ EPD Late Packets’ EPD t  Small p value: smaller Dp, fewer early packets, longer flooding delay, less energy cost => Energy-Sensitive  Large p value: larger Dp, more early packets, shorter flooding delay, more energy cost => Time-Sensitive Decision Making Process

15. Decision Conflict Resolution  The selection of flooding senders Only a subset of neighbors are considered as a node’s flooding packet senders. Flooding senders have a good enough link quality between each other. l th

16. Decision Conflict Resolution  Link-quality based back-off scheme Better link quality, higher chance to send first Further avoids collision when two nodes can hear each other and make the same decision Further saves energy since the node with the best link quality has the highest chance to send

17. Simulation  Simulation Setup Randomly generated network, 200~1000 nodes Randomly generated working schedules Duty cycle from 1%~20% 300m × 300m field The simulation results are based on 10 network topologies and 1000 flooding packets for each topology.

18. Simulation  Baseline 1: optimal performance bounds Delay optimal: collision-free pure flooding Energy optimal: tree-based solution  Baseline 2: improved traditional flooding Two techniques are added to avoid collisions:  Link-quality based back-off scheme  p-persistent back-off scheme

19. Simulation

26. Implementation and Evaluation  Test-bed Implementation 30 MicaZ nodes form a 4-hop network Randomly generated working schedules Duty cycle from 1% to 5%

27. Implementation and Evaluation Flooding delay vs. Duty CycleEnergy Cost vs. Duty Cycle

28. Implementation and Evaluation Ratio of Opportunistically Early Packets Hop Count 1 Hop Count 2 Hop Count 3 Hop Count 4

29. Conclusions  The flooding process in low-duty-cycle networks consists of a number of unicasts. This feature calls for a new solution.  Opportunistically early packets are forwarded outside the energy-optimal tree to reduce the flooding delay.  Late packets are not forwarded to reduce energy cost.  Evaluation reveals this approaches both energy- and delay- optimal bounds.

30. Thanks~~