Energy Efficient Data Collection In
Distributed Sensor Environments
Qi Han, Sharad Mehrotra, Nalini Venkatasubramanian
{qhan, sharad, nalini} @ics.uci.edu
QUASAR Project
University of California, Irvine
School. of Information & Computer Science
Ubiquitous Sensor Environments
• Generational advances to
computing infrastructure
Habitat Monitoring Battlefield Monitoring
Earthquake
Monitoring
Sensor
Networks
Medical Condition
Monitoring
• Continuous monitoring and
recording of physical world and
its phenomena
– limitless possibilities
• New challenges
– limited bandwidth & energy
– highly dynamic systems
Oceanographic
current monitoring
Video
Surveillance
• System architectures are due
for an overhaul
– at all levels of the system
networks, OS, middleware,
databases, applications
Traffic
Congestion
Detection
Target Tracking
– sensors will be everywhere
Intrusion Detection
2
Quasar (Quality Aware Sensing
Architecture)
• Hierarchical architecture
– data flows from producers to
server to clients periodically
– queries flow the other way:
client
server server cache
and archive
producer & its cache
QUERY FLOW
DATA FLOW
client cache
• if client cache does not suffice:
– query routed to appropriate
server
• if server cache does not suffice:
– access current data at producer
– this is a logical architecture
• producers could also be clients
• a server may be a base station or
a (more) powerful sensor node
• servers might themselves be
hierarchically organized
• the hierarchy might evolve over
time
3
Quasar: Observations & Approach
• Applications can tolerate errors in sensor data
– applications may not require exact answers:
• small errors in location during tracking or error in answer to query result
may be OK
– data cannot be precise due to measurement errors, transmission
delays, etc.
• Communication is the dominant cost
– limited wireless bandwidth, source of major energy drain
• Quasar Approach
– exploit application error tolerance to reduce communication between
producer and server and/or to conserve energy
– two approaches
• Minimize resource usage given quality constraints
• Maximize quality given resource constraints
4
This Paper…
• Explore data collection protocols for sensor
environments that exploits the natural tradeoff
between application quality and energy
consumption at the sensors
– Consider a series of sensor models that progressively
expose increasing number of power saving states
– For each of the sensor models considered, develop
quality-aware data collection mechanisms that ensure
quality requirements of the queries while minimizing the
resource consumption
5
Data Collection Framework
query Q1
(A1,D)
source-initiated update
query Qm
(Am,D)
…
consumer-initiated request
sensor si
consumer-initiated update
i=[li,ui]
Imprecise data
representation
• If query quality tolerance satisfied at server
– Answer query at the server
• Else
– Probe the sensor
– Sensor guaranteed to respond within a bounded time D
6
Abstract Sensor States
radio mode
1-radio node
2-radio node
Tx on, Rx off
Tx on, Rx on
sensor state
active (a)
Tx off, Rx on
listening (l)
Tx off, Rx off
sleeping (s)
7
Problem Statement
•
•
Objective: minimize sensor energy consumption in the process of
answering all queries
– Given user queries with varying accuracy constraints and latency bound
Formally stated:
Esu  Psu
minimize E   Ecu  Pcu
 Eextra
S.T.
•
Issues
(energy consumptio n due to source - initiated updates)
(energy consumptio n due to consumer - initiated update)
(energy consumptio n due to idling in different states)
(1)a i  Ai
(accuracy constraint is met)
(2)t i  D
(latency bound is met)
– How to maintain the precision range r for each sensor
• Larger r increases possibility of expensive probes
• Small r wastes communication due to source-initiated updates
– When to transition between sensor states
• Powering down might not be optimal if we have to power up immediately
• Powering down may increases query response time
8
Our Approaches
• We solve the energy optimization problem by
solving two sub-problems
– Optimize energy consumption by adjusting range size
under the assumption that the state transition is fixed
– Optimize energy consumption by adapting sensor states
while assuming that the precision range for sensor is
fixed
• Progressively expose increasing number of sensor
power saving states
–
–
–
–
AA: Always Active
AL: Active-Listening
AS: Active-Listening
ALS: Active-Listening-Sleeping
9
The AL(Active-Listening) model
Upon first source-initiated
update or probe
listening
active
Ta after processing last
source-initiated update or probe
10
Analysis of the AL Model
normalized
sensor energy
consumption:
Eal  f ( Pa , Pl , Psu , Pcu )
sensor state
transition probabilities
steady state
probabilities: Pa , Pl
probabilities of
source- or consumerinitiated updates:
Psu , Pcu  f (r )
r : interval size
Pa : prob. of sensor being ' active'
Pl : prob. of sensor being ' listening'
Psu : prob. of source - initiated updates
Pcu : prob. of consumer - initiated updates
re-write sensor energy
consumption equation:
Eal  f (r )
sensor energy consumption
is minimized when
Pcu
2
Psu
11
Range Size Adjustment for the AA/AL Model
• Optimal range can be realized by maintaining the
P
probability ratio Pcu
su
• Can be done at the sensor
• Assuming that  is the ratio of consumer-initiated
update probability to source-initiated update
probability:
for source-initiated update:
with probability min{,1}, set r’= r(1+);
for consumer-initiated update:
with probability min{1/,1}, set r’=r/(1+ );
12
The AS Model (Active-Sleeping)
Upon first source-initiated update
or after Ts without traffic
sleeping
active
Ta after processing last
source- or consumer-initiated update
13
The ALS Model (Active-Listening-Sleeping)
sleeping
After Tl
without traffic
listening
Upon first source-initiated update
or after Ts
Upon first source-initiated
update or probe
active
Ta after processing last
source-initiated update or probe
14
Range Size Adjustment
for the AS/ALS Model
• Not possible to express the ratio  in terms of other
parameters
– Need to monitor parameters such as K1, K2 etc.
• Sensor side
– Keep track of the number of state transitions of the last k
updates
– Piggyback the probability of state transitions with the Kth
update
• Server side
– Keep track of the number of sensor-initiated updates and
probes of the last k updates
– Upon receiving the Kth update from the sensor
• Compute the optimal precision range r
• Inform the sensor about the new r
15
Adaptive Sensor State Management
• Consider the AS model for derivation of optimal Ta to
minimize energy consumption
– Assuming (t) is the probability of receiving a request at time
instant t, the expected energy consumption for a single silent
period is
Ta
Ta Ts
0
Ta
E    (t ) PCa tdt  
 (t )[ PCaTa  PCs (t  Ta )  Esa ]dt
– E is minimized when Ta=0 if requests are uniformly
distributed in interval [0, Ta+Ts].
• In practice, learn (t) at runtime and select Ta
adaptively
– Choose a window size w in advance
– Keep track of the last w silent period lengths and summarizes
this information in a histogram
– Periodically use the histogram to generate a new Ta
16
Adaptive State Management (Cont.)
• ci : the number of silent periods for bin i among the last w
silent periods
• estimate  by the distribution which generates a silent
period of length ti with probability ci/w
• Ta is chosen to be the value tm that minimizes the energy
consumption as follows:
n c
m1 c j

j
min   PCa  t j    PCa  tm  PCs  (t j  tm )  Esa 
tm
j m w
 j 1 w



c1
c0
cn-1
c2
bin 0
t0
bin 1
bin n-1
bin 2
t1
t2
t3
……
tn-1
tn=Ta+Ts
17
Performance Study
• Modeling sensor
– Sensor values:
• uniformly from the range [-150, 150];
• perform a random walk in one dimension: every second, the
values either increases or decreases by an amount sampled
uniformly from [0.5,1.5].
• Modeling queries
– query arrival times at the server are Poisson distributed
• mean inter-arrival time = 2 seconds.
– each query is accompanied by an accuracy constraint A
• A=uniform( Aavg(1- Avar ), Aavg(1+ Avar ))
• Aavg =20 (average accuracy constraint)
• Avar=1 (accuracy constraint variation)
18
System Performance Comparison of
Proposed Sensor Models
Sensor Energy Consumption Comparison
800
16
normalized sensor energy
consumption(uJ)
average query respone time (us)
Query Response Time Comparison
700
600
500
400
300
200
100
0
14
12
10
8
6
4
2
0
AA
AL
AS
ALS
AA
AL
AS
ALS
19
Impact of Ta adaptation
on System Performance
840
Impact of Ta Selection on Sensor Energy Consumption
normalized sensor energy
consumption(uJ)
average query response time(us)
Impact of Ta Selection on Query Response Time
820
800
780
760
740
720
700
static Ta(0)
adaptive Ta
9
8
7
6
5
4
3
2
1
0
static Ta(0)
adaptive Ta
20
Impact of Range Size Adaptation
on System Performance
Impact of Range Size Adjustment
on Query Response Time
normalized sensor
energy consumption(uJ)
Impact of Range Size Adjustment
on Sensor Energy Consumption
average query
response time (ms)
2500
2000
1500
1000
500
0
fixed(0)
average accuracy
constraint
adaptive
adjustment
fixed(large)
0.05
0.04
0.03
0.02
0.01
0
fixed(0)
average accuracy
constraint
adaptive
adjustment
fixed(large)
21
Conclusions
• Explored the tradeoff between sensor data
accuracy and energy consumption for sensor data
collection in distributed sensor environments
• Both theoretical analysis and experimental results
validated the effectiveness of our approaches
– The AS model consumes the least amount of sensor
energy
– Our proposed strategies of adaptive sensor state
transition reduce energy consumption to a great extent
– Optimized range size adjustment works effectively with
corresponding sensor models and saves more energy than
using static range or instantaneous values
22