Data Compression Algorithms for Energy-Constrained Devices inDelay Tolerant Networks

Christopher SadlerMargaret Martonosi

Princeton University

2

1

10

100

1000

10000

100000

1000000

10000000

CC2420 CC1000 XTend

Radio

MSP

430

Clo

ck C

ycle

s fo

rSa

me

Ener

gy a

sO

ne B

yte

Tran

smitt

ed

Why do we care about Compression in Sensor Networks?

Short Range

125 m

Med. Range

300 m

Long Range

15 km

Success = Energy Savings

Energy = Compute Energy + Transmit Energy

~2 Million!

~32,000

~4,000

3

0

20

40

60

80

100

120

140

100% 75% 50% 25%

Percentage of Packets Received Correctly

Thou

sand

s of

MSP

430

Clo

ck

Cyc

les

for S

ame

Ener

gy a

s O

ne

Byt

e Tr

ansm

itted

Why do we care about Compression in Sensor Networks?

Unreliability:Retransmission Extra energy costEasier to amortize original energy cost of compression

Medium Range Radio (CC1000)

~128,000!

~32,000

4

Why do we care about Compression in Sensor Networks?Source

Local Energy Tradeoff:Transmit all data

vs.Compress dataStore dataTransmit compressed data

Sink

Downstream Energy Tradeoff:Relay all data

vs.Relay compressed data

Downstream Energy Tradeoff:Relay all data

vs.Relay compressed data

Savings Accumulate with Hop Count

5

This work Lossless compression algorithms tailored to static

and mobile sensors

Generally applicable for WSNs Aggregation Spatial-temporal correlation

Implementation and evaluation on the real data set Great Duck Island Monitoring, ZebraNet, …

6

Outline

Design Criteria and LZW Compression What we want in Sensor Compression? How do we adapt LZW to Sensors?

Using Compression to Conserve Energy Conclusions

7

Sensor Network Compression: Energy Savings for Everyone Need a general purpose, lossless compression

algorithm that can work across the design space

ZebraNet (Princeton): Outdoors, Mobile

Great Duck Island (UCB): Outdoors, Stationary

SensorScope (EPFL): Indoors, Stationary

8

Sensor Network Compression: What do we want? Low Transmission Overhead Computationally Simple

Compute energy of compression should not outweigh transmission energy savings

Bounded Memory Footprint To fit in small sensor node memories

Adaptive to general data sets Exploit repetition in general input data streams Work on small blocks of data

9

LZW Compression

LZW is a dictionary-based algorithm which encodes new strings based on previously encountered strings.

AAAABAAABCC (8bits * 11 = 88bits required) Standard Dic

A = 65 B = 66C = 67. . . Z = 90. . . ? = 255

Encoded String Output Stream New Dic entryA 65 AA = 256 AA 65 256 AAA = 257 A 65 256 65 AB = 258 B 65 256 65 66 BA = 259 AAA 65 256 65 66 257 AAAB = 260 B 65 256 65 66 257 66 BC = 261 C 65 256 65 66 257 66 67 CC = 262 C 65 256 65 66 257 66 67 67 66 bits required

10

Outline

Design Criteria and LZW Compression What we want in Sensor Compression? How do we adapt LZW to Sensors?

Using Compression to Conserve Energy Conclusions

11

S-LZW: LZW for Sensor Nodes Dictionary decisions

How large should we make the dictionary? What do we do if the dictionary fills?

Data decisions How much data should we compress at once? Can we shape the dictionary to improve compression? Can we shape the data to make it easier to compress?

12

S-LZW: LZW for Sensor Nodes Dictionary decisions

How large should we make the dictionary? What do we do if the dictionary fills?

Data decisions How much data should we compress at once?

Longer data streams => better compression learning But too long => high retransmit cost when packets dropped

Can we shape the dictionary to improve compression? Can we shape the data to make it easier to compress?

Examined by experiments

13

SENSOR DATA – N BYTES GENERATED OVER TIME

…528 B Block (2 Flash Pages)

COMP. ALGORITHM

COMP. ALGORITHM

COMP. ALGORITHM

Compressed Data

Compressed Data

Compressed Data

Independent groups of 10 or fewer dependent packets

… … …

………

528 B Block (2 Flash Pages)

528 B Block (2 Flash Pages)

S-LZW Idea 1: Data Size

14

S-LZW Idea 2: Mini-Caching

Dictionary Tree Mini-

Cache (N entries)

0 1

Escape Bit

9 Bit Entries10 Bit Entries (Log2 N)+1 Bit Entries

Exploit fine-grained locality even in short sensor data sequences Proposal: Mini-cache to tightly encode MRU entries

Hit => Saves multiple bits. Miss => Costs just 1 extra escape bit

15

S-LZW Idea 3: Data Transforms BWT – Reversible method of transforming

data used in bzip2 and applicable for all data Proposal: Structured Transform - Create a

matrix of readings and transpose it to create runs Simple, but effective

cb 4e 70 62 …d5 4e 46 62 …d8 4e 31 62 …db 4e 2b 62 …... ... ... ...

cb d5 d8 db …4e 4e 4e 4e …70 46 31 2b …62 62 62 62 …... ... ... ...

16

Outline

Design Criteria and LZW Compression Using Compression to Conserve Energy

Local and Downstream Energy The Influence of Unreliable Communications The Effects of Shaping the Data

Conclusions

17

Measurement Methodology:CPU and Radios Evaluation Platform:

TI microcontroller MSP430: 10 kB RAM, 48 kB ROM Off-Chip Flash: 4 Mbit Atmel 3 Radios: Short (CC2420), Medium (CC1000), and Long

(XTend) range 3 Real World Datasets…

SensorScope (SS) – Indoor, Stationary Great Duck Island (GDI) – Outdoor, Stationary ZebraNet (ZNet) – Outdoor, Mobile

… And one Compression Benchmark Geo from the Calgary Corpus (Calgeo)

18

Local Energy Savings

00.2

0.40.60.8

1

1.21.4

SS

GD

I

ZNet

Calgeo

Dataset

Nor

mal

ized

Ene

rgy

00.20.40.60.8

11.21.4

SS

GD

I

ZNet

Calgeo

Dataset

Nor

mal

ized

Ene

rgy

0

0.00

2

0.00

4

0.00

6

0.00

8

0.01

0.01

2

None

SS

GDI

ZNet

Calgeo

CPU

Flas

hR

adio

CC2420(Short Range)

XTend(Long Range)

Data Compressed with S-LZW with Mini-Cache

Model assumes 100% reliability

2.6X Gain

1.2X Gain1.7X

Gain

1% Loss

15+% Loss

Low

er is

bet

ter

19

Downstream EnergySavings

0

1

2

3

4

5

1 2 3 4 5 6 7 8 9 10

Hop Count

Ener

gy S

aved

(J)

0

1

2

3

4

5

1 2 3 4 5 6 7 8 9 10

Ener

gy S

aved

(J)

0

1

2

3

4

5

1 2 3 4 5 6 7 8 9 10

Hop Count

-0.005

0

0.005

0.01

0.015

0.02

1 2 3 4 5 6 7 8 9 10

Hop Count

Ener

gy S

aved

(J)

0

1

2

3

4

5

1 2 3 4 5 6 7 8 9 10

Hop Count

CC2420 (Short Range) XTend (Long Range)

ZNet data, Compressed with S-LZW with Mini-Cache

Model assumes 100% reliability

20

Coping with Unreliability1. Energy savings

increase linearly with hop count

GDI Data, Compressed with S-LZW with Mini-Cache

CC2420 (Short Range)

21

Coping with Unreliability1. Energy savings

increase linearly with hop count

GDI Data, Compressed with S-LZW with Mini-Cache

CC2420 (Short Range)

2. At a 90% success rate, we save energy locally

22

Transforms to Improve Performance BWT – Reversible method of transforming

data through sorting itXTend (Long Range)

We sort with an iterative quicksort algorithm, but this comes at a high computational cost

Normalized against sending the data without compression

3.4X Gain

2.4X Gain

7-8% ImprovementTop – Radio EnergyMiddle – Flash

EnergyBottom – CPU Energy

Model assumes 100% reliability

23

Radio Range?

Number ofHops?

Data Composition?

Number ofHops?

Structured General

Radio Range? S-LZW-MC8-ST~2.3X Savings

RLE-ST~1.9X Savings

S-LZW-MC8-ST~2.4X Savings

Few Many

Short Medium/Long

S-LZW-MC16-BWT~1.8X Savings

S-LZW-MC32~1.4X Savings

S-LZW-MC16-BWT~1.8X Savings

Few Many

Short Medium/Long

Algorithms Summary

MC – Mini-CacheST – Structured Transform

BWT – Unstructured Transform

24

Conclusions Existing techniques (LZW, Cache, BWT …) applied

to new application domains (WSNs) Less novelty and originality Incremental work Upper bound 4 points

Systematic solution with complete considerations Dictionary size, data size, local cache, data transformation

… Substantial merits proved by sound experiments

Different types of radio transceivers (CC2420, CC1000, XTend)

Real-world data set