Synopsis Diffusion for Robust Aggregation in Sensor Networks

1
Synopsis Diffusion for Robust Aggregation in
Sensor Networks

• Suman Nath et. al
• CMU, Intel

2
Aggregation

• In large networks, aggregation queries more
  important than individual sensor readings.
• Aggregates in-network saves energy.
• Popular approach is to create spanning tree
• Drawback is that a single node failure can lead
  to loss of the entire subbranchs data.

3
Stumbling Block

• Fundamental block is tight coupling of
  aggregation and the required routing topology is
  tightly couples.
• This limits the advantage that can be achieved if
  we were to use robust routing, like multi path
  routing.
• Multi path routing has a drawback i.e duplication
  of data.

4
Synopsis Diffusion

• Synopsis Diffusion energy efficient multipath
  routing smart algorithms to avoid double
  counting
• ODI Order and Duplicate Insensitive synopses

5
Papers Contributions

• A Novel Aggregation Framework
• Properties to characterize ODI Synopsis
• Better Aggregation Topologies
• Adaptive Ring Topology
• Example Aggregates
• Median
• Performance Evaluation

6
Synopsis Diffusion

• Synopsis Generation SG
• Takes sensor reading and generates a synopsis
• Synopsis Fusion SF
• Takes two synopses and generates a new one.
• Synopsis Evaluation SE
• Translates a synopsis into final answer.
• Three Magic Functions to take care of all woes

7
Phases of SD

• Distribution Phase
• Aggregate query is flooded through the network
• Aggregation Phase
• Each node uses SG to convert local data to local
  synopsis and then uses SF to merge two synopsis
  to create a new one. The query initiator uses the
  SE to generate the final result.

8
Synopsis Diffusion

• Synopsis Label

9
ODI Correctness

• Algorithm is ODI correct if
• SF And SG Order and Duplicate

10
Properties
11
Semi Lattice
12
Error Bounds
13
Additional Examples

• Uniform sample of sensor readings.

14
Uniform Sampling (Cont)

15
Adapting the Topology

• Ring Topology
• Adaptive Ring Topology
• Nodes moves up or down in the rings dependent
  upon the messages it overhears.

16
Evaluation

• 20 ft X 20 ft
• 600 sensors
• TAG Simulator
• Aggregation Scheme
• TAG ( standard tree based)
• TAG2 (value splitting between two parents)
• RINGS
• Adaptive RINGS
• FLOOD

17
Impact of sensor density

18
Sensor density on power consumption

19
Correlated Node failure

20
Conclusions

• First paper to use this approach.
• Acclaims four easy properties to solve
  everything.
• My thoughts are it is very hard to come up with
  good SF and SG functions.