Adaptive Sleep Discipline for Energy Conservation in Sensor Networks

1
Adaptive Sleep Discipline forEnergy Conservation
in Sensor Networks Jana van Greunen, Alvise
Bonivento alvise, janavg_at_eecs.berkeley.edu
2
Power Aware Protocols

• Power aware protocol design
• Limited complexity applications
• Power aware routing
• Optimized Physical layer smart use of
  broadcasting, beaconing.

• Randomized Algorithms
• Balance work load
• Increased performance with node density

• Adaptive Algorithms
• Adaptive to traffic variations and node density

• Distributed Algorithms
• Reduce overhead
• Increase responsiveness
• Challenge
• Optimize parameter evaluation in a distributed
  fashion

3
Sleeping Discipline

• SLEEP IF YOU CAN
• Maintain connectivity
• If the node is not necessary, goes to sleep and
  saves power
• For how long should the node be allowed to sleep
  ?

• Given
• Loss rate
• Delay constraint
• Data generation requirement

Find optimal sleeping discipline

• Our Solution
• Adaptive
• Traffic node density
• Random
• Exponentially distributed sleeping times.
• Avoid phase synchronization.

• PRESENT SOLUTIONS
• Mainly fixed sleeping solutions
• Simple to design
• Not adaptive
• Not scalable
• Leads to inefficiency

4
Routing Model
One controller for the whole region Region
divided in a grid of square blocks Geographical
routing Forward packet to the first available
node in an adjacent block closer to the
controller
Sleeping regions

• Nodes know (block granularity)
• Their own location
• The controller location

C
B
No information of the node density
A
5
SLEEPING DISCIPLINE Algorithm

• Evaluate traffic decide if the actual sleeping
  times are adequate for the delay constraint
• Additive increase activity if the actual sleeping
  average was not enough or multiplicative decrease
  activity if the activity level is too high
• Receive packets and possibly forward them to the
  next hop
• Generate random exponentially distributed random
  variable with the updated activity (as explained
  in 3) that will represent the next sleeping time.
• Go to sleep if you have no packets left else Stay
  awake until all packets are forwarded

6
Estimation
Crucial step for effective adaptiveness !!
PASTA Estimation
How to compute optimal µ for each region ?
Constraint the per-hop delay
Exponential distribution is memory-less. Can
estimate time of last node wakeup (i.e. before
first packet arrival)
Assume the sleeping time is a random variable
exponentially distributed with parameter µ
7
Results

• Algorithm works for different topologies
• Different size
• Different density
• AIMD ensures good fairness in most of the cases

Fairness for different topologies
Fairness in a large dense network
Fairness with different disciplines
8
Results
End-to-end delay increases linearly with the
number of hops Packets dont experience
additional delay as they aggregate
End to end delay vs number of hops
Per hop delay density more important than size
Average per hop delay
9
Comparison Results
Adaptive sleeping time wins !!!
Exponentially distributed sleeping discipline has
a better load distribution !!
Exponentially distributed sleeping time
Deterministic sleeping time
10
Conclusions Future Work

• Conclusions
• Outperformed existing solution !!
• Adaptation has a great impact
• Random disciplines better than deterministic
• Fairness ensured with AIMD
• Future Work
• Extend analysis to more complex power models
• So far only two states considered (on/off)
• Algorithm for optimal region sizing