Decentralized Scattering of Wake-up Times in Wireless Sensor Networks Amy L. Murphy ITC-IRST,...

16
Decentralized Scattering of Wake-up Times in Wireless Sensor Networks Amy L. Murphy ITC-IRST, Trento, Italy joint work with Alessandro Giusti, Politecnico di Milano, Italy Gian Pietro Picco, University of Trento, Italy

Transcript of Decentralized Scattering of Wake-up Times in Wireless Sensor Networks Amy L. Murphy ITC-IRST,...

Decentralized Scattering of Wake-up Times in Wireless Sensor Networks

Amy L. MurphyITC-IRST, Trento, Italy

joint work with

Alessandro Giusti, Politecnico di Milano, ItalyGian Pietro Picco, University of Trento, Italy

A common solution to save energy on sensing devices is to periodically turn them on and off

Energy saved while turned off lead to clear gains in system lifetime

We explore duty cycling at the application level, cycling communication and/or sensing– Spreading out awake times, NOT synchronizing them

Results in lifetime increase, but at a cost…

Energy Management with Duty Cycling

times1

epochawake interval

wakeup time

Cost of Duty Cycling: response time

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

Random Wakeup Scattered Wakeup

4 transmissionsrequired

2 transmissionsrequired

Scenario– Distributed nodes– Mobile base station– BS queries nearby nodes– Queries repeat until a node

responds– Communication duty cycled:

respond only when turned on

Scenario– Distributed sensing nodes– Sensors detect events within a

given radius, only when sensors are active

– Duty cycle the sensing capability

– Note: long wakeup intervals

Cost of Duty Cycling: event coverage

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

AC

B

Random Wakeup

AC

B

A

Scattered Wakeup

A C B

“Less” likely todetect events

“More” likely todetect events

Wake-up Scattering: distributed protocol

All nodes discover– Their wake up time, WC

– Wake up time of the node before them, Wprev

– Wake up time of the node after them, Wnext

Calculate their new target wake up time– Wc’=(Wprev+Wnext)/2 * – Move toward this new

wakeup time in the next epoch

Key Properties– Process is entirely localized– All nodes execute in parallel– No central coordination

ABC

WC

WnextWprev

WC’

0.00

0.05

0.10

0.15

0.20

0.25

init 1 2 3 4 5 10 20 50Iteration Number

Average Response Delay (fraction of E)

Rapid convergence to goodscattering (2-3 rounds) Increasing

awake intervals

A=0.50

A=0.01

Scattering to reduce response delayA

vera

ge R

espo

nse

Del

ay (

frac

tion

of E

)

Scattering Iteration Numberinit

A=0.15

A=0.10

Same responsedelay

No scattering.Long awake time.

After scattering.33% shorter awake time.

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

0.20

init 1 2 3 4 5 10 20 50Iteration Number

Average Response Delay (fraction of E)

Effectiveness of scattering for response delay: Different network densities

Ave

rage

Res

pons

e D

elay

(fr

actio

n of

E)

Scattering Iteration Numberinit

Range=100, 8.4 neighbors

Range=110, 10.1 neighborsSame response

delay

No scattering.Large radio range.

After scattering.10% smaller radio range.

0.4

0.5

0.6

0.7

0.8

0.9

1

init 1 2 3 4 5 10 20 50Iteration Number

Sensing Coverage

Scattering to increase sensing coverage

Goal: increase percentage of events detected by scattering the awake intervals of the sensors themselves

Results are similar to those for response delay– Details in the paper

After scattering,same coverage.

20% smaller awake interval.

Visualization

http://www.elet.polimi.it/upload/giusti/scattering

Sensing radius

Node ON

Node OFF

Overlappingsensing

Node

Pairwisecommunication

Scattering & Latency in Tree-based data collection on WSN

Many WSN are used to collect data at a central location by constructing an overlay tree along which data flows

Goal: low latency for data from source to sink

In terms of wake-up times, the parent should wake up after the child to receive its data

C

X

A B

Scattering and Tree Formation

Consider a simple tree, and the wake-up times from the perspective of X– Well scattered, but X needs to send data to C

– Would be better if C wakes up immediately after X, not A

Scattering never changes the sequence of wakeup times– We introduce jumping, a simple mechanism

that allows reordering in the sequence of wakeup times

– After jumping, additional scattering is required

Jumping enabled if the Wnext is a child

– With some probability, select next wakeup time between Wnext and Wnextnext

A

XB

C

C

X

A B

0

1

2

3

4

5

6

init scattering jumping

Time To Root

A=0.01 A=0.05 A=0.1 A=0.15

Reducing Latency with Jumping

Time to root not significantly affected by scattering alone

Reducing gap betweenwake-ups reduces latency.

However….

Initial

Scattering

Jumping

4 different awake intervals

Jumping to place parentsafter children results in significant improvement

Tim

e to

Roo

t(l

ow

er is

bet

ter)

Waving: reduce gap between wakeup times

Tradeoff between “tree” and “coverage”

Reducing space betweenwake-ups reduces latency.

However….

…benefits of scattering reduced.Result: lower event coverage

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

init scattering jumping

0

1

2

3

4

5

6

init scattering jumping

Tim

e to

Roo

t(l

ow

er is

be

tter

)P

erce

nt C

over

age

(hig

he

r is

be

tter

)

Visualization: Tree

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

http://www.elet.polimi.it/upload/giusti/scattering

Discussion

Combining wakeup scattering for coverage and jumping to achieve good, tree-based data collection yields a promising complete solution

Wakeup Scattering is fully decentralized– Wake up times are determined based on local information– Epochs need not be synchronized across nodes

Simple algorithm yields significant results– Response delay: same as random wakeup times with 33% longer

awake interval– Event coverage: same as random with 20% longer awake interval– Tree: scattering + jumping improve over random from 25 to 45%

http://www.elet.polimi.it/upload/giusti/scattering

Future Directions

Modify the awake interval to meet the application needs, e.g., increased coverage

Exploit signal strength to approximate distance between sensors– Close sensors should have “more scattered” awake

times

Combine jumping to improve solutions for response delay and coverage– Avoid local minima in scattering solution

Consider applying scattering at the MAC layer

http://www.elet.polimi.it/upload/giusti/scattering