Distributing Queries Over Low Power Sensor Networks

1
Distributing Queries Over Low Power Sensor
Networks Sam Madden, Robert Szewczyk,
Michael Franklin, Wei Hong, Joe Hellerstein, and
David Culler
Focus Hierarchical Aggregation
Wireless Sensor Networks
Palm DevicesLinux

• Aggregation natural in sensornets
• The big picture typically interesting
• Aggregation can smooth noise and loss
• UDAs to do signal processing
• Provides data reduction
• Power/Network Reduction in-network aggregation
• Hierarchical version of parallel aggregation
• Tricky design space
• Metrics power cost and answer quality
• Variables topology-selection, value-routing
  scheme, other tricks
• Dynamic environment requires adaptive schemes

Smart Dust MotesTinyOS

• A spectrum of devices
• Varying degrees of power and network constraints
• This demo Mica and TinyOS
• Focus on many/tiny
• Toward MEMS Smart Dust
• Off-the-shelf HW for now Berkeley Mica Mote
• Wireless, single-ported, ad-hoc network
• Spanning-tree communication through root

Performance in Tiny SensorNets
A Query Language for Sensors
Aggregation and NW Optimization

• Power consumption
• Communication gtgt Computation
• METRIC radio wake time
• Send gt Receive
• METRIC messages generated
• Bandwidth Constraints
• Internal gtgt External
• Volume gtgt surface area
• Result Quality
• Noisy sensors
• Discrete sampling of continuous phenomena
• Lossy communication channel

• Continuous queries with streaming, periodic
  results
• UDAs and UDFs
• Currently compiled-in
• Mote Virtual Machine (Mate) under development
• Periodic nature allows for
• Scheduling of communication and sleep
• Simple semantics for combining multi-hop readings
• Clearly other alternatives here
• E.g. sequence/timeseries/temporal query languages

• An expanded taxonomy of aggregates
• State
• Duplicate sensitivity
• Montonicity
• Exemplary vs. Summary
• Effects on
• Value Routing
• Snooping and Suppression
• Caching and Presumption
• Hypothesis Testing
• Collapsing of the NW and QP layers!

SELECT ltaggsgt, ltattrsgt WHERE
ltpredsgtGROUP BY ltexprgt HAVING ltpredsgtEPOCH
DURATION ltconstantgt
TinyDB Software On Motes

• 4200 lines of C Code
• Runs on Mica Motes with light and temperature
  sensors, magnetometers and accelerometers
• 4Mhz Atmel Processor
• 4KB RAM, 40kBit radio, 512K EEPROM, 128K Flash
• Ad-hoc queries
• Java UI
• Split-pane display
• Topology visualization
• Applications
• Environmental, military
• NW Monitoring!