1. Manjunath Doddavenkatappa, Chan Mun Choon and Ben Leong National University of Singapore Splash: Fast Data Dissemination with Constructive Interference in Wireless Sensor Networks

3. 3

4. A Fundamental Service: Data/Program Dissemination 4

5. A dissemination protocol is required throughout the life of a sensor application 5 Dissemination completion time is critical

6. Completion time for existing protocols is still in the order of minutes 6 Culprit is Contention Resolution

7. SPLASH 7 Constructive Interference [Glossy’2011] Channel Diversity [PIP’2010] Eliminate the need for contention

8. 8

9. SPLASH Transmission Density Diversity Tree Pipelining Channel Cycling Opportunistic Overhearing XOR Coding 9

10. 10

11. Cycle 1 P1 Transmitting Receiving Idling P1 Cycle 2 P1 Cycle 3 P2 P1 P2 Cycle 4 P1 11

12. SPLASH Transmission Density Diversity Tree Pipelining Channel Cycling Opportunistic Overhearing XOR Coding 12

13. SPLASH Channel Cycling XOR Coding 13 Transmission Density Diversity Tree Pipelining Opportunistic Overhearing

14. Constructive Interference is not Scalable [Wang et al. INFOCOM’12] The problem is more severe in practice 14

15. 15 Decreasing trend is not always true Constructive Interference is not Scalable

16. Effect of the Capture Effect Our results suggest that an increase in no. of transmitters can also improve reliability 16

17. Splash: Transmission Density Diversity First round: only non-leaf nodes forward Typically, more than 50% of the nodes in a tree are leaf nodes [Manjunath et al. RTSS’11] Nodes benefit from low transmission density Second round: all nodes transmit Nodes benefit from high transmission density by exploiting capture effect or sender diversity [Rahul et al. SIGCOMM’10] 17

18. SPLASH Transmission Density Diversity Tree Pipelining Channel Cycling Opportunistic Overhearing XOR Coding 18

19. Cycle 2 P1 X Cycle 3 P2 P1 19

20. P1 Cycle 3 P2 P1 20 Cycle 2 P1 X Packet reception chance is doubled

21. SPLASH Transmission Density Diversity Tree Pipelining Channel Cycling Opportunistic Overhearing XOR Coding 21

22. CH1 CH2 CH3 CH4 CH3 CH2 CH1 Round 1 Round 2 22

23. SPLASH Transmission Density Diversity Tree Pipelining Channel Cycling Opportunistic Overhearing XOR Coding 23

24. Splash: XOR Coding After two rounds, more than 50% of nodes received most but not full object Third round: every transmission is a XOR sum Probability that a packet transmission is useful is increased 24

25. A node in Splash has six chances to receive a packet on up to six different channels Transmission Density Diversity Opportunistic Overhearing Channel Cycling XOR Coding 25 Splash: Summary of Its Three Dissemination Rounds Three Rounds of Dissemination

26. Splash: Local Recovery If any missing data is recovered locally Neighbor querying and data downloading over CSMA/CA Fact that about 90% of nodes have the full object makes local recovery practical 26

28. Experimental Setup 139 TelosB nodes Indriya Testbed 90 Tmotesky nodes Twist Testbed

29. Comparison Splash (Contiki) DelugeT2 (TinyOS) Other Protocols Deluge (Contiki)

30. SplashDelugeT2 Tree [no.] Size [hops] R1 [%] R2 [%] R3 [%] N R3-100% [%] R lr [%] T Splash [sec] T DelugeT2+CTP [sec] T DelugeT2GI [sec] 1584.5497.2398.4791.3010022.491300924 2686.5296.9198.5892.0310022.61286160 3776.6894.6297.8086.2310023.18209286 4788.0296.1297.7892.7510023.74218158 5976.9793.6596.6981.8810023.86649180 6776.7395.2798.1689.8610025.98610160 7780.7593.5196.9889.1310026.25365379 8783.5794.4396.0187.6810026.89377277 9582.4695.2697.4785.5110028.09676313 10884.2894.9296.7086.2310028.39550216 Average82.0595.1997.4688.2610025.15524305.3 Summary of Results on Indriya (32 KB dissemination) 25.15s 524s 305s /21 /12 30

31. Summary of Results on Twist (Splash: 32 KB, Deluge: 2KB) SplashDeluge Tree [no.] Size [hops] R1 [%] R2 [%] R3 [%] N R3-100% [%] R lr [%] T Splash (32KB) [sec] T Deluge (2KB) [sec] 1490.5897.0999.2294.3810020.07356.60 2481.0894.7099.3192.1310020.19431.48 3486.5396.1998.0091.0110022.79351.67 4478.6494.1098.1284.0910023.37518.19 5481.4293.9597.9889.8910023.41467.00 6478.0493.5596.8285.3910026.66439.81 7483.9095.1897.5489.8910026.79345.28 8483.7093.6496.4584.2710027.32388.68 9681.5893.3597.0285.3910027.45484.10 10580.7893.0997.1185.3910029.25397.59 Average82.6294.4897.7688.1810024.73418.04 24.73s 418s 31

32. Comparison to Other Approaches (Baseline: DelugeT2) ProtocolNo. of nodes File size [KB] Reduction factor MNP (2005)10051.21 MC-Deluge (2005)2524.31.6 Rateless Deluge (2008)200.71.47 ReXOR (2011)1641.53 ECD (2011)25101.44 MT-Deluge (2011)200.72.42 Splash1393221 32

33. Evaluation of Individual Techniques Channel Cycling Local Recovery XOR coding Opportunistic Overhearing Transmission Density Diversity 33

34. Evaluation of Individual Techniques 34 XOR Coding: increases percentage of nodes having the full object from 37% to 88% Opportunistic overhearing: decreases dissemination time by 26%

35. Evaluation of Individual Techniques 35 Channel Cycling: decreases dissemination time by 25% Transmission Density Diversity: 39% and 18% of nodes benefit from low and high transmission densities respectively

36. Local Recovery 36

37. Local Recovery 8s for 98.8% of 30 KB 12.4s for 1.2% of 30 KB 37

38. Conclusion We designed and implemented Splash, a fast data dissemination protocol Key factors are constructive interference, channel diversity and techniques for reliability Splash reduces completion time by an order of magnitude 38

39. Thank you 39

40. References 40 Some of the images in these slides are downloaded from Google Images and Kalyan Varma’s websitehttp://kalyanvarma.net /http://kalyanvarma.net /