Combs, Needles, and Haystacks: Balancing Push and Pull for Information Discovery Xin Liu Computer Science Dept. University of California, Davis Collaborators: Qingfeng Huang & Ying Zhang, PARC Presented by Chien-Liang Fok on March 4, 2004 for CSE730
11/4/2004 ACM Sensys2 Objective Simple, reliable, and efficient on-demand information discovery mechanisms
11/4/2004 ACM Sensys3 Where are the tanks?
11/4/2004 ACM Sensys4 Pull-based Strategy
11/4/2004 ACM Sensys5 Pull-based Cont’d
11/4/2004 ACM Sensys6 Push-based Strategy
11/4/2004 ACM Sensys7 Comb-Needle Structure
11/4/2004 ACM Sensys8 Assumptions Events: Anywhere & Anytime Queries: Anywhere & Anytime Global discovery-type One shot Network: Uniform Examples: Firefighters query information in the field Surveillance Sensor nodes know their locations
11/4/2004 ACM Sensys9 When an Event Happens
11/4/2004 ACM Sensys10 When a Query is Generated Event Query Event
11/4/2004 ACM Sensys11 Tuning Comb-Needle
11/4/2004 ACM Sensys12 Query Freq. < Event Freq.
11/4/2004 ACM Sensys13 Query Freq. < Event Freq.
11/4/2004 ACM Sensys14 Reverse Comb When query frequency > event frequency
11/4/2004 ACM Sensys15 The Spectrum of Push and Pull PullPush Global pull +Local push Global push +Local pull Push & Pull Inter-spike spacing increases Reverse comb Relative query frequency increases
11/4/2004 ACM Sensys16 Mid-term Review Basic idea: balancing push and pull Preview: Reliability Random network An adaptive scheme
11/4/2004 ACM Sensys17 Strategies for Improving Reliability Local enhancement Interleaved mesh (transient failures) Routing update (permanent failures) Spatial diversity Correlated failures Enhance and balance query success rate at different geo-locations Two-level redundancy scheme l=2s
11/4/2004 ACM Sensys18 Spatial Diversity Query x Event Diversify query spatially using green arrows
11/4/2004 ACM Sensys19 Random Network Constrained geographical flooding Needles and combs have certain widths
11/4/2004 ACM Sensys20 Simulation Using Prowler Transmission model: Reception model: Threshold MAC layer: Simulates Berkeley Motes’ CSMA Use Default radio model: σ a =0.45, σ b =0.02, p error =0.05, =0.1
11/4/2004 ACM Sensys21 Two Experiments 1. What is the optimal spacing of the comb & needle length given F q and F e ? 2. What is the robustness of the protocol in a really sparse network?
11/4/2004 ACM Sensys22 Experiment 1 Results l=1, s=3 optimal l optimal ~
11/4/2004 ACM Sensys23 Experiment 2 Results Wider the CGF width More Reliable More Energy
11/4/2004 ACM Sensys24 Adaptive Scheme Comb granularity depends on the query and event frequencies Nodes estimate the query and event frequencies to guess s Important to match needle length and inter-spike spacing Allow asymmetric needle length Comb rotates Load balancing Broadcast information of current inter-spike spacing
11/4/2004 ACM Sensys25 Simulation 20x20 regular grid Communication cost: hop counts No node failure Adaptive scheme
11/4/2004 ACM Sensys26 Event & Query Frequencies
11/4/2004 ACM Sensys27 Tracking the Ideal Inter-Spike Spacing
11/4/2004 ACM Sensys28 Simulation Results Gain depends on the query and event frequencies Even if needle length < inter-spike spacing, there is a chance of success. Tradeoff between success ratio and cost 99.33% success ratio and 99.64% power consumption compared to the ideal case
11/4/2004 ACM Sensys29 Summary Adapt to system changes Can be applied in hierarchical structures PullPush Global pull +Local push Global push +Local pull Push & Pull Relative query frequency increases
11/4/2004 ACM Sensys30 Future work Further study on random networks Building a “comb-needle-like” structure without location information Integrated with data aggregation and compression Comprehensive models for communication costs