Taming the Underlying Challenges of Reliable Routing in Sensor Networks Alec Woo, Terence Tong, and David Culler UC Berkeley and Intel Research Berkeley A Cross-Layer Perspective of Routing How to get from A to B? Underlying question: what are the ways to get from A to B? not given vary over time Each layer is a distributed, local process. Combine Select Good Routes Neighbor management keep the good ones Discover & characterize connectivity Study and understand global properties End-to-end success rate Routing topology over time Stability November 5 th, 2003 SenSys 2003 2 Underlying Connectivity in Reality 3 regions and transitional region is large Effective Clear Deployment: Communication range = effective region Nodes: Discover connectivity = link estimation Hear many nodes in transitional region How to define a neighbor? Transitional many November 5 th, 2003 SenSys 2003 3 Roadmap Underlying process Link Quality Estimation Neighborhood Management Cross-layer routing study Tree-based Routing November 5 th, 2003 SenSys 2003 4 Accurate A Good Link Estimator Passive Estimation Link sequence number snooping Estimate inbound reception quality Agile yet stable Small memory footprint Simple November 5 th, 2003 SenSys 2003 5 Key issue Cannot infer losses until next packet reception Solution With a minimum data rate, infer losses based on time Asymmetric links Require outbound transmission quality estimation Exchange reception quality over local broadcast November 5 th, 2003 SenSys 2003 6 1
Link Estimator Study Study 7 estimators by tuning to yield the same error bound Results WMEWMA(T, α) Estimator Stable, simple, constant memory footprint Compute success rate over non-overlapping window (T) Average over an EWMA(α) Key: 10% error requires at least 100 packets to settle Limits rate of adaptation November 5 th, 2003 SenSys 2003 7 Neighbors in Transitional Hear Many potential neighbors Few good nodes (blue) Potential neighbors > available table-size Get in Neighbor Cannot tell which neighbor is good Get out General solution: down-sample to suppress gray nodes maintain frequent nodes November 5 th, 2003 SenSys 2003 8 Management Techniques Cache Replacement Policy FIFO, LRU (LRH), Clock Database Frequency estimation of data streams FREQUENCY (Manku et al.) November 5 th, 2003 SenSys 2003 9 Details Insert (when) Down-sample rate adapt to # of neighbors Rate = Size / Neighbors Reinforce if in table Cache hit (FIFO, LRH, Clock) Node s Counter++ (Frequency) Evict (which) Counter--, zero entries are replaceable (Freq) If all Counter > 0, drop insertion Cache policies evict for each insertion November 5 th, 2003 SenSys 2003 10 Key Results Fixed-size table as cell density increases 1st 2nd 3rd # Good neighbors > size Freq always maintains 50% or more good neighbors in table Roadmap Underlying process Link Quality Estimation Neighborhood Management Cross-layer routing study Tree-based Routing 40 Number of Potential Neighbors November 5 th, 2003 SenSys 2003 11 November 5 th, 2003 SenSys 2003 12 2
Distributed Tree Building Distributed distance-vector protocols Operate over link estimator and neighbor management Send periodic route messages Carry cost to tree root Piggyback link estimations Hop-Count? Neighbor? Link Quality? Hop-Count Hear neighbor s cost and store in table Select minimum cost neighbor for routing Route damping What should be the cost metric? November 5 th, 2003 SenSys 2003 13 Shortest hop with no estimation (SP) Shortest hop with threshold SP(%) Hard threshold November 5 th, 2003 SenSys 2003 14 Non-threshold based Cost Metrics Path Reliability Product of link quality along the entire path Minimum Transmission (MT) Cost is based on link quality Link estimator provides 1 p forward p hop reverse Cost = E[total number of trans.] November 5 th, 2003 SenSys 2003 15 Evaluation Roadmap Key observations: Hop distribution, end-to-end success, stability Graph analysis Large 80x80 grid SP, SP(%), MT Rule out SP because of poor reliability Packet-level simulation 10x10 grid, (max 2 retrans./hop) Broadcast and DSDV (periodic route selection) Neighbor table management (Freq -> MTTM) Empirical (Mica Motes) Small 5x10 grid and 30-node random placement SP(%), MT with large enough table max 2 retrans./hop, deliberate congestion High Level Low Level November 5 th, 2003 SenSys 2003 16 Graph Analysis Key Results Hop-Distribution and Reliability to BS Simulation Key Results End-to-end Hop-Count Success Stability Distribution vs. Distance November 5 th, 2003 SenSys 2003 17 November 5 th, 2003 SenSys 2003 18 3
Empirical Study Restudy connectivity vs. distance Put nodes at end of effective region (~ worst case) 8 feet Study SP(70%), SP(40%), MT Key observations: SP(70%) fails SP(40%) fails Hard threshold fails under congestion Link quality drops under traffic November 5 th, 2003 SenSys 2003 19 Different Empirical Key Results from simulations! End-to-end Hop-Count Success Distribution vs. Distance Effective is 8 feet November 5 th, 2003 SenSys 2003 20 Average Hop-Count Contour Plot 30-node network Congestion and Stability Topology Stability # Route Changes Per 5 Route Messages Link Estimation % Time (s) November 5 th, 2003 SenSys 2003 21 November 5 th, 2003 SenSys 2003 22 Conclusion Connectivity: 3 regions (need to measure) Deployment: node spacing Routing as 3 on-line processes Discover/maintain varying connectivity (WMEWMA) Settling time limits adaptation rate Maintain a set of good neighbors regardless of cell density (Frequency) Hard thresholds are problematic MT (No predefine threshold) Interactions among these processes Hop distribution, end-to-end success, stability Backup Slides November 5 th, 2003 SenSys 2003 23 4
Routing Architecture Routing Cost: Actual vs. Est. Send originated data message Application Timer Originating Queue Send route update message Cycle detected choose other parent Parent Selection Run parent selection and send route message periodically Forward Queue Cycle Detection Forwarding Data message message Filter discard non data packet discard duplicate packet Estimator All message sniff and estimate Neighbor Route message save information Management All Messages November 5 th, 2003 SenSys 2003 25 November 5 th, 2003 SenSys 2003 26 Channel Utilization Contour MT Topology Movie November 5 th, 2003 SenSys 2003 27 November 5 th, 2003 SenSys 2003 28 5