Matching Data Dissemination Algorithms to Application Requirements

1
Matching Data Dissemination Algorithms to
Application Requirements

• 2004. 11. 2
• John Heideman, Fabio Silva and Deborah Estrin
• SenSys 2003
• Kisuk Kweon

2
Contents

• Introduction
• Summary of Diffusion algorithms
• Two-phase pull diffusion
• Geographically scoped data with GEAR
• Push diffusion
• One-phase pull diffusion
• Push vs. two-phase pull diffusion
• Systematic Comparison
• Algorithm performance with different numbers of
  sources and sinks
• Varying the number of sinks with one-phase pull
• Cost of adding more skins with push
• Using geographic information
• Conclusions

3
Introduction

• What is the sensor networks ?
• Composed of a large number of sensor nodes that
  are densely deployed inside the phenomenon

4
Introduction

• Application involvement in sensor-network
  communication
• Application-specific, data-centric communication
  protocols
• Reduce communications cost
• Selection of which communications algorithms best
  match applications
• New algorithms supporting the diffusion API
• Directed diffusion implied both a interface and
  routing implementation
• Push diffusion
• Optimized for many receivers but few senders
• One-phase pull diffusion
• Optimized for many senders but few receivers

5
Summary of Diffusion Algorithms

• The key abstraction of the diffusion API
• Data is identified by a set of attributes
• Data producers generate data by publishing
• Data consumers subscribe to data
• The business of the diffusion implementation
• Insure that data travels from publisher to
  subscriber efficiently

6
Two-phase pull diffusion

• Initial work with diffusion used two-phase pull
• Interests are flooded to find any data sources
• Nodes establish gradients
• The source sends exploratory data by flooding
• The sink reinforces its preferred neighbor
• Nodes can also generate negative reinforcement
• Gradient are managed as soft-state

7
Geographically scoped data with GEAR

• Geographic and Energy-Aware Routing (GEAR)
• Extends diffusion when node locations and
  geographic queries
• Replaces network-wide communication with
  geographically constrained communication
• Added to two-phase pull diffusion
• Interest is flooded within interesting region
• Exploratory data is sent only on gradients set up
  by interests

8
Push diffusion

• A different class of applications
• Applications with many sources and sinks
• Sources produces data only occasionally
• Push diffusion
• The roles of the source and sink are reversed
• Exploratory data is sent throughout the network
  without interested gradients
• The rest process is the same as two-phase pull

Source
Source
Sink
Sink
Exploratory data
Send data
9
One-phase pull diffusion

• Two-phase pull to eliminate one of its phase of
  flooding
• Subscribers send interest establishing gradients
• Does not sends first data as exploratory
• Sends data only on the preferred gradient
• Does not require reinforcement
• Two disadvantage
• Assumes symmetric communication between nodes
• Requires interest message to carry a flow-id

10
Push vs. two-phase pull diffusion

• The cross-subscription application
• Push is designed for the case when there are many
  sinks, but few nodes generating data
• 7 Sensoria WINSng 2.0 nodes
• All cross-subscribed to each other
• Generates readings every 5s and changes state
  every minute

11
Systematic Comparison

• Methodology
• The EmStar simulation/emulation
• environment
• 60 nodes randomly in a square
• area 50m
• Interest interval set to 30s
• Exploratory data interval at 90s

12
Algorithm performance with different numbers of
sources and sinks
13
Varying the number of sinks with one-phase pull
14
Varying the number of sinks with one-phase pull
15
Cost of adding more sinks with push
16
Using geographic information
17
Conclusions

• A single algorithm performed well with some
  applications but poorly with others
• Matching algorithm to application is important
• Push works best with many sinks and few active
  sources
• One-phase pull works best with many sources and
  few sinks
• With geographic optimization we can serve
  complementary applications performance