Slides (ppt) - Advanced Data Management Technologies Laboratory

Energy Efficient Data Management
in Sensor Networks
Sanjay K Madria
Web and Wireless Computing Lab (W2C)
Department of Computer Science, Missouri
University of Science and Technology, Rolla,
MO
www.mst.edu/~csweddb
Introduction
•
Energy challenges in WSN motivates to devise
algorithms which require least energy to make the
sensors network last longer.
•
Power consumption due to excessive wireless
communication and computation, therefore, need
to minimize both.
•
Example: Data Aggregation helps in minimizing the
number of wireless transmissions in a multi hop
communication scenario.
•
Counter Example: Secure data aggregation can
drain energy saved by data aggregation and
introduce delays and therefore, need better energy
efficient solutions.
OBJECTIVE




Adaptive watermarking-like techniques to provide
confidentiality and integrity verification of high speed
sensor data streams.
Tailored towards energy efficiency to enhance
lifetime, minimal computational overhead to enhance
availability, while simultaneously being adaptive in
order to meet application demands on desired security
and compression levels.
Homomorphic encryption and additive digital
signature schemes (using public key cryptography) for
providing confidentiality of sensor data during
aggregation in WSNs
Algorithm to allow aggregate encrypted data in
Wireless Sensor Networks and a digital signature
scheme to preserve data integrity.
Implementation on Two Platforms
MICA2
TELOSB
Operation
Time Taken Energy
(ms)
Consumed
(mJ)
Time
Taken
(ms)
Energy
Consumed
(mJ)
Encryption
Sign
4406.207
105.75
10573.63
57.09
2477.036
59.44
5890
31.80
Addition of ciphertext
335.2
8.04
406.59
2.19
Addition of signatures
0.09
0.002
0.2
0.001
Addition of Public
keys
157.67
3.78
200
1.08
Energy Efficeint Compression for WSN:
TinyPack Numeric - Temporal Locality
Temperature Readings
18.05
18
17.95
17.9
17.85
17.8
17.75
17.7
17.65
17.6
1
6
11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
Distribution of Values
Distribution of Delta Values
16
40
14
35
12
30
10
8
6
25
20
15
4
2
17
.7
5
17
.7
7
17
.7
9
17
.8
1
17
.8
3
17
.8
5
17
.8
7
17
.8
9
17
.9
1
17
.9
3
17
.9
5
17
.9
7
17
.9
9
18
.0
1
0
10
5
0
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
TinyPack Numeric
Methods

TP-Initial



TP-Dynamic Frequencies



Static delta codes
Fast, efficient, good compression ratio
Huffman style delta codes over window
More processing and RAM, better
compression
TP-Running Statistics


Approximate frequencies by data
statistics
Similar compression, less RAM
TinyPack Numeric
Results (compared with two existing methods)
Compression
30%
25%
20%
15%
Real Life Datasets
Aircraft
Health
N-CET
Triangulate
N-CET
Track
Intel Labs
S-LZW
LEC
TP-Init
TP-DF
TP-RS
Great Duck
Island
10%
ZebraNet
compressed size
35%
TinyPack Numeric
Results (compared with two existing methods)
Energy Usage
Latency
60%
percent of time required to send
uncompressed
percent of energy required to send
uncompressed
60%
55%
50%
45%
40%
35%
50%
40%
30%
20%
10%
0%
30%
S-LZW
S-LZW
LEC
TP-Init
TP-DF
TP-RS
Processing Time
LEC
TP-Init
Send Time
TP-DF
TP-RS
Wait Time
TinyPack XML
Results (compared with three existing methods)
Compressed Size
Delay Tolerant (collect data, then compress)
20%
15%
10%
XMill
xmlppm
TinyPack
paq
5%
0%
Target
Intercept
SpeakerID
SNAResult
Real Life Datasets
Compressed Size
Real Time (compress while collecting)
80%
60%
40%
XMill
xmlppm
paq
TinyPack
20%
0%
Target
Intercept
SpeakerID SNAResult
Real Life Datasets
PAQ
requires
prohibitive
amounts of
memory and
time and is
included as a
benchmark
Decentralizing Compression
Implementation for Real Time Sensor Networks







Spatio-temporal correlation
Data can be approximated to increase
correlation
Group sensors with similar data –
Distinct grops
Choose a base signal
Transmit ratio signals (scalar multiple
of base signal)
Ratio signals have high compressibility
Compressed Sensing
Decentralizing Compression
Implementation for Real Time Sensor Networks

Decentralization



Define groups at sinks and cluster
heads
Choose base signal with leader
selection
Compression

Lossless


Compress ratio signals using TinyPack
Lossy
Set maximum tolerated error
 Compress with TinyPack and run length
encoding
