Sensor network: storage and query. Overview. TAG Introduction. Overview. Device Capabilities

Transcription

1 Sensor network: storage and query TAG: A Tiny Aggregation Service for Ad- Hoc Sensor Networks Samuel Madden UC Berkeley with Michael Franklin, Joseph Hellerstein, and Wei Hong Z. Morley Mao, Winter Slides based on Sam Madden stalk Z. Morley Mao, Winter TAG Introduction Overview What is a sensor network? Programming sensor nets is hard! Declarative queries are easy - Tiny Aggregation (TAG): In-network processing via declarative queries - Simulation & Results Eample: Vehicle tracking application: weeks for students Vehicle tracking query: took minutes to write, worked just as well! SELECT MAX(mag) WHERE mag > thresh EPOCH DURATION 6ms Z. Morley Mao, Winter Z. Morley Mao, Winter Overview Device Capabilities - Simulation & Results Mica Motes - 8bit, Mhz processor Roughly a PC AT - kbit CSMA radio - KB RAM, 8KB flash, KB EEPROM - TinyOS based Variety of other, similar platforms eist - UCLA WINS, Medusa, Princeton ZebraNet, SmartIts Z. Morley Mao, Winter Z. Morley Mao, Winter 6

2 Sensor Net Sample Apps Metric: Communication Habitat Monitoring: Storm petrels on great duck island, microclimates on James Reserve. Vehicle detection: sensors along a road, collect data about passing vehicles. Earthquake monitoring in shaketest sites. Traditional monitoring apparatus. Z. Morley Mao, Winter 7 Lifetime from one pair of AA batteries - - days at full power - 6 months at % duty cycle Communication dominates cost - < few ms to compute - ms to send message Our metric: communication! Current (ma) Time v. Current Draw During Query Processing Snoozing Processing Processing and Listening Transmitting... Time (s) Z. Morley Mao, Winter 8 Communication In Sensor Nets Overview Radio communication has high link-level losses - typically about m Ad-hoc neighbor discovery B A C - Optimizations & Results Tree-based routing D E F Z. Morley Mao, Winter 9 Z. Morley Mao, Winter Declarative Queries for Sensor Networks Aggregation Queries Eamples: SELECT nodeid, light WHERE light > EPOCH DURATION s Epoch Nodeid Light Z. Morley Mao, Winter 89 Sensors Temp Accel Sound Epoch AVG(sound) SELECTAVG(sound) EPOCH DURATION s Epoch roomno AVG(sound) 6 SELECTroomNo, AVG(sound) GROUP BY roomno HAVING AVG(sound) > EPOCH DURATION s 7 Rooms w/ sound > Z. Morley Mao, Winter

3 Overview TAG - Optimizations & Results In-network processing of aggregates - Common data analysis operation Aka gather operation or reduction in programming - Communication reducing Operator dependent benefit - Across nodes during same epoch Eploit semantics improve efficiency! Z. Morley Mao, Winter Z. Morley Mao, Winter Query Propagation Basic Aggregation SELECT COUNT(*) Epoch Comm. Slot In each epoch: - Each node samples local sensors once - Generates partial state record (PSR) local readings readings from children - Outputs PSR during its comm. slot. At end of epoch, PSR for whole network output at root (In paper: pipelining, grouping) Z. Morley Mao, Winter Z. Morley Mao, Winter 6 Slot Slot Z. Morley Mao, Winter 7 Z. Morley Mao, Winter 8

4 Slot Slot Z. Morley Mao, Winter 9 Z. Morley Mao, Winter Aggregation Framework Z. Morley Mao, Winter Slot As in etensible databases, we support any aggregation function conforming to: Agg n ={f init, f merge, f evaluate } f init {a } <a > F merge {<a >,<a >} <a > F evaluate {<a >} aggregate value (Merge associative, commutative!) Eample: Average AVG init {v} <v,> AVG merge {<S, C >, <S, C >} < S + S, C + C > AVG evaluate {<S, C>} S/C Partial State Record (PSR) Z. Morley Mao, Winter Types of Aggregates Taonomy of Aggregates SQL supports MIN, MAX, SUM, COUNT, AVERAGE TAG insight: classify aggregates according to various functional properties - Yields a general set of optimizations that can automatically be applied Any function can be computed via TAG In network benefit for many operations - E.g. Standard deviation, top/bottom N, spatial union/intersection, histograms, etc. - Compactness of PSR Property Partial State Duplicate Sensitivity Eemplary vs. Summary Eamples MEDIAN : unbounded, MAX : record MIN : dup. insensitive, AVG : dup. sensitive MAX : eemplary COUNT: summary Effects Effectiveness of TAG Routing Redundancy Applicability of Sampling, Effect of Loss Monotonic COUNT : monotonic AVG : non-monotonic Hypothesis Testing, Snooping Z. Morley Mao, Winter Z. Morley Mao, Winter

5 TAG Advantages Simulation Environment Communication Reduction - Important for power and contention Continuous stream of results - Smooth transient faults across epochs Lots of optimizations - Via operator semantics Evaluated via simulation Coarse grained event based simulator - Sensors arranged on a grid - Two communication models Lossless: All neighbors hear all messages Lossy: Messages lost with probability that increases with distance Z. Morley Mao, Winter Z. Morley Mao, Winter 6 Benefit of In-Network Processing Optimization: Channel Sharing ( Snooping ) Simulation Results Nodes Grid Depth = ~ Neighbors = ~ Total Bytes Xmitted Total Bytes Xmitted vs. Aggregation Function Some aggregates require dramatically more state! EXTERNAL MAX AVERAGE COUNT MEDIAN Aggregation Function Z. Morley Mao, Winter 7 Insight: Shared channel enables optimizations Suppress messages that won t affect aggregate - E.g., MAX - Applies to all eemplary, monotonic aggregates Z. Morley Mao, Winter 8 Optimization: Hypothesis Testing Eperiment: Hypothesis Testing Insight: Guess from root can be used for suppression - E.g. MIN < - Works for monotonic & eemplary aggregates Also summary, if imprecision allowed How is hypothesis computed? - Blind or statistically informed guess - Observation over network subset Messages / Epoch Messages/ Epoch vs. Network Diameter (SELECT MAX(attr), R(attr) = [,]) No Guess Guess = Guess = 9 Snooping Uniform Value Distribution Dense Packing Ideal Communication Z. Morley Mao, Winter 9 Network Diameter Z. Morley Mao, Winter

6 Optimization: Use Multiple Parents Multiple Parents Results For duplicate insensitive aggregates Or aggregates that can be epressed as a linear combination of parts - Send (part of) aggregate to all parents In just one message, via broadcast - Decreases variance B A C / / Z. Morley Mao, Winter Better than No previous Splitting analysis epected! Critical Losses aren t independent! Link! Insight: spreads data over many links Avg. COUNT Benefit of Result Splitting (COUNT query) Z. Morley Mao, Winter 8 6 With Splitting ( nodes, lossy radio model, 6 parents per node) Splitting No Splitting Summary TAG enables in-network declarative query processing - State dependent communication benefit - Transparent optimization via taonomy Hypothesis Testing, Parent Sharing Declarative queries are the right interface for data collection in sensor nets! - Easier to program and more efficient for vast majority of users Z. Morley Mao, Winter 6