1. 1 st Oct 2013 1 CorLayer: A Transparent Link Correlation Layer for Energy Efficient Broadcast Shuai Wang, Song Min Kim, Yunhuai Liu, Guang Tan, and Tian He University of Minnesota MobiCom 2013

2. The Need for Broadcast Operation 2 University of Minnesota Shuai Wang @ MobiCom’ 13 Code Dissemination Global Time Sync Routing Discovery Data Collection Wireless communication essentially occurs in a broadcast medium.

3. Multi-path Routing Opportunistic Forwarding Network Coding 3 University of Minnesota Shuai Wang @ MobiCom’ 13 The Need for Broadcast Operation Advanced designs exploit the benefit from broadcast nature.

4. Motivation Despite the fact that wireless communication essentially occurs in a broadcast medium with concurrent receptions Existing research predominately examine separate statistics for individual links (channel) or path: ETX, PPR, LQI, RSSI Little research has been done to investigate the joint statistics involving concurrent wireless links (e.g. broadcast) Because of the legacy assumption of link independence link independence 4 University of Minnesota Shuai Wang @ MobiCom’ 13

5. Legacy Assumption It is assumed that wireless reception among concurrent links are independent due to multipath induced fading. 5 N1N1N1N1 N2N2N2N2 S i.e., Packet loss at N2 is independent of packet loss at N1. University of Minnesota Shuai Wang @ MobiCom’ 13

6. Unfortunately…. Legacy assumption no longer holds well because packet loss due to the coexistence of wireless networks 6 University of Minnesota Shuai Wang @ MobiCom’ 13

7. 7 The co-existence of ZigBee and Wi-Fi University of Minnesota Shuai Wang @ MobiCom’ 13

8. The co-existence of ZigBee and Wi-Fi 8 Wireless spectrum becomes crowded: 802.11b, 802.11g, and 802.15.4 all use the 2.4 GHz ISM band. Interference becomes the major cause of pack loss instead of fading 25dB difference University of Minnesota 1119 Shuai Wang @ MobiCom’ 13

9. Explosive Growth of Wi-Fi 9 1100% Wi-Fi Hotspots in U.S. University of Minnesota Shuai Wang @ MobiCom’ 13

10. Increasing Cross-Network Interference Two receivers' PRR The concurrent noise increase Interference leads to correlated packet loss: 10 University of Minnesota Shuai Wang @ MobiCom’ 13

11. 11 Furthermore, Correlated Shadow Fading Closely located Closely located devices may suffer correlated lose since wireless signals suffer shadow fading when obstacles appear in the propagation path of the radio waves. University of Minnesota Shuai Wang @ MobiCom’ 13

12. Furthermore, Correlated Shadow Fading 12 University of Minnesota Two receivers' PRR The concurrent RSSI reduction CloselyLocated Appearance of Obstacles University of Minnesota Shuai Wang @ MobiCom’ 13

13. Synthetic Independent Trace Empirical Trace 13 Wireless links are correlated! University of Minnesota 1 Source node 9 Receivers 100 Packets University of Minnesota Shuai Wang @ MobiCom’ 13

14. How Link Correlation Affects Broadcast？ (a) Negative Correlated: (b) Positive Correlated: In order to accurately estimate the broadcast performance, we MUST consider link correlation. Link quality: 0.8 # of packets need to be retransmitted: 4 Link quality: 0.7 # of packets need to be retransmitted: 3 14 University of Minnesota Shuai Wang @ MobiCom’ 13

15. The expected number of transmissions : Theoretical Analysis 15 Transmissions due to Link Quality Reduced transmissions by Link Correlation University of Minnesota Shuai Wang @ MobiCom’ 13 : the probability that all nodes in K(u) successfully receive a packet.. K i (u) is a subset of N(u) with size i, where N(u) is node u’s one-hop neighbor set.

16. 16 University of Minnesota Shuai Wang @ MobiCom’ 13 Special Case – when links are independent: The Property of Property 1: Property 2: The higher the link correlation - The fewer the transmissions -

17. Link Blacklisting for Better Correlation 17 The average number of transmissions before blacklisting is mainly concentrated around 4.5 and it's 2.4 after blacklisting. University of Minnesota Shuai Wang @ MobiCom’ 13 2.4 4.54.54.54.5 Empirical Study : Blacklisting leads to a significant reduction in transmission!

18. CorLayer: Goals Design a supporting layer by blacklisting low correlated links to help upper layer protocols save transmissions. Goals: Design a supporting layer by blacklisting low correlated links to help upper layer protocols save transmissions. 18 University of Minnesota Shuai Wang @ MobiCom’ 13 Neighbor Discovery CorLayer Broadcast Protocols Original Physical Topology Updated Logical Topology

19. 19 CorLayer: Challenges How to guarantee the network connectivity when blacklisting is executed? How to guarantee the network connectivity when blacklisting is executed? A localized light-weight algorithm for connectivity check. A localized light-weight algorithm for connectivity check. How to blacklist links thus the updated topology can benefit the upper layer broadcast protocols? How to blacklist links thus the updated topology can benefit the upper layer broadcast protocols? Assess the cost of covering one-hop neighbors, taking link correlation into consideration. Assess the cost of covering one-hop neighbors, taking link correlation into consideration. University of Minnesota Shuai Wang @ MobiCom’ 13

20. CorLayer: Design – Connectivity Check 20 Key Idea – link blacklisting requires the existence of an alternative path. W UV Asynchronously Blacklisting – two-phase locking is used to avoid a race condition. University of Minnesota Shuai Wang @ MobiCom’ 13

21. CorLayer: Design – Link Blacklisting University of Minnesota Shuai Wang @ MobiCom’ 13 Key Idea - Triangular Blacklisting Rule: blacklisting a link if the source node could take fewer transmissions via an alternative path. UV W x z y 21

22. 22 CorLayer: Design – Link Blacklisting University of Minnesota Shuai Wang @ MobiCom’ 13 Key Idea - Triangular Blacklisting Rule: blacklisting a link if the source node could take fewer transmissions via an alternative path. W UV Direct Broadcast Cost for 1st Hop Cost for 2nd Hop UV W x UV W x

23. Evaluation Testbed Settings 23 University of Minnesota Shuai Wang @ MobiCom’ 13 PlatformLocationEnvironmentNodes/APs MICAzUMNLab36/5 TelosBSIATOffice30/8 GreenOrbsTRIMPSOutdoor20/0 Physical SizeDegreeChannelPower 8m*2.5m7~23Ch16-25dBm 18m*13m6~21Ch16, Ch26-25dBm 15m*5m4~13Ch16-25,-19.2dBm 23

24. Supported Protocols (1/2) 24 University of Minnesota Shuai Wang @ MobiCom’ 13 Integrated Protocols: Integrated Protocols: I.Tree based: 1). S-Tree: A. Juttner et al. Mobile Networks and Application’05 2). C-Tree: K. Alzoubi et al. HICSS’02 II.Cluster based: 3). Cluster: J. Wu et al. Wireless Communication and Mobile Computing’03 4). Intermediate: J. Wu et al. Telecommunication Systems’01 5). Clustering: I. Stojmenovic et al. TPDS’02 6). P-Clustering: T. J. Kwon et al. SIGCOMM’02 III.Multiple point relay: 7). MPR: A. Qayyum et al. HICSS’02 8 – 9). Min-id MPR, MPRCDS: C. Adjih et al. INRIA-Rapport’02

25. Supported Protocols (2/2) 25 University of Minnesota Shuai Wang @ MobiCom’ 13 Integrated Protocols: Integrated Protocols: IV.Pruning based: 10-11). SP, DP: H. Lim et al. Computer Communications Journal’01 12-13). PDP, TDP: W. Lou et al. TMC’02 14). RNG: J. Cartigny et al. IJFCS’03 V.Location based: 15). CCH: M. T. Sun et al. CS-NMC’05 VI.Network Coding: 16). COPE: S. Katti et al. SIGCOMM’06 16). CODEB: E. L. Li et al. INFOCOM’07 Evaluation Metrics: Evaluation Metrics: The total number of transmissions needed to deliver one packet to all the nodes in the network. The total number of transmissions needed to deliver one packet to all the nodes in the network. Extensive evaluation with 16 protocols run on 3 testbeds!

26. Evaluation Main Performance Results 26 University of Minnesota Shuai Wang @ MobiCom’ 13 38% 48% 52% 49% 36% 39% On average, CorLayer reduces transmissions by 47%!

27. Evaluation Impact of blacklisting rules Impact of blacklisting rules R_: Random Blacklisting; R_: Random Blacklisting; WL_: Worst Link Blacklisting; WL_: Worst Link Blacklisting; CorLayer_: Our Design; CorLayer_: Our Design; 27 R_: CorLayer Saves “R_”50% Transmissions! WL_: CorLayer Saves “WL_” 20% Transmissions! MPRClusterPruning Network Coding University of Minnesota Shuai Wang @ MobiCom’ 13

28. Conclusion We have presented CorLayer, a link correlation-based layer that enhances the energy efficiency of reliable broadcasting. We have presented CorLayer, a link correlation-based layer that enhances the energy efficiency of reliable broadcasting. We integrated CorLayer transparently with sixteen state-of- the-art broadcast algorithms and evaluated the design on three real-world multi-hop testbeds. We integrated CorLayer transparently with sixteen state-of- the-art broadcast algorithms and evaluated the design on three real-world multi-hop testbeds. The results indicate that with CorLayer, reliable broadcast avoids unnecessary transmissions caused by wireless links that are less positively correlated. The results indicate that with CorLayer, reliable broadcast avoids unnecessary transmissions caused by wireless links that are less positively correlated. 28 University of Minnesota Shuai Wang @ MobiCom’ 13

29. Thank you! Q&A Q&A 29 University of Minnesota Shuai Wang @ MobiCom’ 13