1. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign When Pipelines Meet Fountain: Fast Data Dissemination in Wireless Sensor Networks Presenter: Huajie Shao (hshao5) Author: Wan Du, Jansen Christian Liando, Huanle Zhang, Mo Li Nanyang Technological University, Singapore 1

2. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Outlines 1.Background 2.Motivation 3.Pando protocol 4.Evaluation 5.Conclusion 1.Background 2.Motivation 3.Pando protocol 4.Evaluation 5.Conclusion 2

3. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Background Volcano [G. Werner-Allen et al., EWSN’ 05, OSDI’06] Wireless sensor networks What is the Data dissemination? Deliver a bulk of data to all the nodes in a network. (e.g., modify codes or an application updating profile) What is the Data dissemination? Deliver a bulk of data to all the nodes in a network. (e.g., modify codes or an application updating profile) 3

4. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Traditional Protocols Deluge Sprinkler Splash 2004 2007 2013 Hop by hop with acknowledgement overhead and low spatial reuse Limit spatial reuse, without considering constructive interference Limit spatial reuse, without considering constructive interference Contention-based protocol Long tail time 4

5. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Splash: Constructive interference + Pipelining Channel 1Channel 2 Channel 3Channel 4 One packet is forwarded simultaneously by all nodes at a same layer which interfere constructively with each other Disseminate data three rounds Disseminate data three rounds Same data object twice Same data object twice 500 XOR Encoded packets 500 XOR Encoded packets 5

6. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Long tail problem Fig 2: The packet reception performance of every node after each round in Splash. 6

7. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Splash: long tail problem >80% of total time duplicate packets 7

8. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Pando Protocol Channel 1Channel 2Channel 3Channel 4 Fountain codes: contains novel information of the data object. Avoid duplicate retransmission Avoid duplicate retransmission Radio-driven coding scheme idle intervals of CPU Deterministic time. idle intervals of CPU Deterministic time. Silence-based feedback scheme Acknowledgement of every node 8

9. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign 1) Fountain codes X1 X2 X3 X4 Y1 Y2 Y3 Y4 Y5 Y6 X1 X2 X3 X4 Recovered Packets Original Packets Encoded Packets Individual packets Gaussian Elimination Number of packets ACK Received Packets Y1 Y3 Y4 Y6 Y5 Y2 9

10. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign 2) Radio-driven coding scheme R1 T1R2 T2 D1 D2 RnTn Dn μCμC Radio 1688μs Accumulative Gaussian Elimination 10

11. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Gaussian Elimination 1) Row operations: Swapping two rows 2) Multiplying a row by a non-zero number, 3) Adding a multiple of one row to another row. Using these operations a matrix can always be transformed into an upper triangular matrix 11

12. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign 3) Silence-based feedback T1T2 R20T20 R2T1 R20 T2 R1 R2T1 R1 R19 T18 R18 T19 R19 T17... R17 T18 R18 T17 T19 R19 T18 T1 T3 R3T3 R3T2 R2 T2 R18 Source Layer 1 Layer 2 Layer 3 Original data: 16 packets; Successful decoding: 17 encoded packets. Generate the encoded packets by its own. Fig. 5 : Dissemination process of a 16-packet data object in Pando 12

13. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Improve Feedback Scheme False negative of channel silence False negative of channel silence False positive of channel silence False positive of channel silence Detect M consecutive silent slots 13

14. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign 4) Packet-level adaptation of channel diversity and network density Figure 6: Packet-level channel allocation 14

15. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Implementation—state machine 15

16. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Indriya: 99 TelosB sensor motes Flocklab: 31 Tmote Sky sensor motes The packet size is set to 64 bytes Data object: 32 kBytes Baselines: Deluge and Splash Four metrics: reliability, dissemination time, energy consumption and memory cost. Evaluation -Testbeds 16

17. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Evaluation: dissemination time Note that, the authors encounter technical problems for the Deluge protocol. 17

18. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Evaluation: dissemination time 18

19. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Reliability of individual node Figure 8, Reliability of individual nodes achieved by Pando and Splash on Indriya and Flocklab. 19

20. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Evaluation – Dissemination time on Flocklab 20

21. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Dissemination time on Flocklab 15 s 29s 21

22. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Dissemination time on Indriya 12 s 30s 22

23. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Compared to others Table 3: Reduction factor of dissemination time over Deluge achieved by the existing data dissemination protocols in wireless sensor networks 23

24. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Conclusion This paper presented Pando, a contention-free data dissemination protocol for wireless sensor networks. Designed a radio-driven coding scheme and adopt Fountain encoding to effectively solve long-tail problem Designed a novel silence based feedback approach in the one-way pipelines to timely acknowledge the successful recovery of the data object and stop dissemination process. Applied packet-level adaptation of channel diversity and network density to boost the dissemination efficiency. The results show that Pando outperformed other baseline protocols This paper presented Pando, a contention-free data dissemination protocol for wireless sensor networks. Designed a radio-driven coding scheme and adopt Fountain encoding to effectively solve long-tail problem Designed a novel silence based feedback approach in the one-way pipelines to timely acknowledge the successful recovery of the data object and stop dissemination process. Applied packet-level adaptation of channel diversity and network density to boost the dissemination efficiency. The results show that Pando outperformed other baseline protocols 24

25. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign Thank you very much ！ Questions? 25

26. Huajie Shao @ University of Illinois. Urbana-Champaign Huajie Shao @ University of Illinois. Urbana-Champaign References Du, Wan, et al. "When Pipelines Meet Fountain: Fast Data Dissemination in Wireless Sensor Networks." Proceedings of the 13th ACM Conference on Embedded Networked Sensor Systems. ACM, 2015. http://www.ntu.edu.sg/home/limo/ https://en.wikipedia.org/wiki/Gaussian_elimination 26