Predictability and Efficiency in Predictability and Efficiency in - - PowerPoint PPT Presentation

Predictability and Efficiency in Predictability and Efficiency in Wireless Sensor Networks

Lothar Thiele

Jan Beutel Federico Ferrari, Matthias Keller, Roman Lim, Andreas Meier, Clemens Moser, MustafaYuecel, Matthias Woehrle

P di t bilit d ffi i Predictability and efficiency

influences the whole life cycle, and is integrated into all abstraction layers and is integrated into all abstraction layers.

Overview

• Sensor Networks – A Critical Review
• Predictability and Efficiency

– Testing and Verification – Protocols – Energy Scavenging

• PermaSense

3

Wireless Sensor Networks Visions

1991 1999 2000 1996 2003 2001 2004

Visions

1991 1999 2000 1996 2003 2001 2004 Ubiquitous Vision Scale Free Networks Smart Dust Directed Diffusion COTS Dust PicoRadio Wireless Paintable Terminodes

Applications

2004 2000 2003 2001 Overlay Computing Military Surveillance Argo Shooter Localization James Reserve ZebraNet Sensor Webs Duck Island

4

Argo – Global Ocean Observation Strategy

• Global array of temperature/salinity profiling floats

– Satellite data relay to data centers on shore – Operational since 2000 – Developed and maintained mainly by oceanographers

5

Wireless Sensor Networks Visions

1991 1999 2000 1996 2003 2001 2004

Visions

1991 1999 2000 1996 2003 2001 2004 Ubiquitous Vision Scale Free Networks Smart Dust Directed Diffusion COTS Dust PicoRadio Wireless Paintable Terminodes

Applications

2004 2000 2003 2001 Overlay Computing

WSN Community

Military Surveillance Argo Shooter Localization James Reserve

Prototypes, Experiments

ZebraNet Sensor Webs Duck Island

and Research Demos

6

“Proof-of-Concept” Deployment Experience

eley] ey] tta, UC Berke ta, UC Berkel [Prabal Dut [Prabal Dutt TU Delft] Berkeley] Langendoen, nn Tolle, UC B [Koen L [Gilman 7

Promises

• Sensor nodes are cheap, so we can have plenty of

p, p y them.

• Nodes may be cheap, but deployment and

y p, p y maintenance is expensive.

• Additional redundant nodes make the system fault
• Additional redundant nodes make the system fault

tolerant automatically.

• More nodes make the system more fragile
• More nodes make the system more fragile.

8

Overview

• Sensor Networks – A Critical Review
• Predictability and Efficiency

– Testing and Verification – Protocols – Energy Scavenging

• PermaSense

9

Today's WSN Design and Development Simulation

TOSSIM

TOSSIM [Levis2003] PowerTOSSIM [Shnayder2004] Avrora [Titzer2005]

Deployment Virtualization and Emulation

EmStar [Ganesan2004]

ale

Deployment

In-network reprogramming

[Levis2004,Hui2004]

C lib ti d V ifi ti EmStar [Ganesan2004]

BEE [Chang2003,Kuusilinna2003]

Sca

id

Calibration and Verification

[Szewczyk2004]

Trial-and-error [Mainwaring2004,

Hemingway2004,Cerpa2001]

Test Grids

moteLab [Werner-Allen2005] Emstar arrays [Cerpa03/04] Dependence on infrastructure

[Szewczyk2004]

Reality

Figure abridged from D. Estrin/J. Elson

Emstar arrays [Cerpa03/04]

Kansei [Dutta2005]

10

DSN: Result Comparison and Research Questions

Platform Runtime Test success Tossim .5 minutes 80% DSN Testbed 11 minutes 0%

• How can we formally

y specify tests (driver, monitor, evaluation)?

• How can we proof the
• How can we proof the

correctness of a test result (synchronization, ti ll d d partially ordered events)?

11

Testbed Example: Deployment-Support Network (DSN) l S k Developer Deployment-Support Network

Temporary, minimal invasive Virtual connections to nodes

Developer Access

Virtual connections to nodes

Reliable, wireless, scalable

Target Sensor Network Network

12

Testbed Example: The Deployment-Support Network Testbed Functions

Remote reprogramming

Remote reprogramming

Extraction of log data Analysis

R i t ti

Regression testing

Target Sensor Network Network

13

DSN Architecture Details

14

DSN Example –Test Case Generation

15

DSN Example – Network Control

16

DSN Example - Sniffer Three-Layer concept

• WSN application
• Spy-nodes listening
• DSN for collecting

captured traffic captured traffic

Evaluation Tool for further l i analysis

17

DSN Example - Sniffer

18

DSN: Physical Characterization Architecture

• Emulating the Environment...

– Temperature Cycle Testing (TCT) Temperature Cycle Testing (TCT)

• ... and Resource Usage

– Different Power Sources: Batteries, rechargeable cells, Different Power Sources: Batteries, rechargeable cells, solar, fixed DC power...

19

DSN: 4 Nodes – Long Term Power Profiling Details

20

DSN: Formal Conformance Testing of Power Traces

21

DSN: Formal Conformance Testing of Power Traces

• Use of timed automata for conformance testing
• Modular modeling of
• Modular modeling of

– Hardware S ft – Software – Testing Environment – Trace

• Complexity reduction by

p y y

– noise removal – abstraction and interval arithmetic

• Testing by reachability analysis of the composed

model (UPPAAL)

22

model (UPPAAL)

DSN: Formal Conformance Testing of Power Traces

• Experiment

– TMOTE Sky (MSP 430 TI CC2420) running TinyOS2 TMOTE Sky (MSP 430, TI CC2420) running TinyOS2 – Harvester application, LPL MAC protocol – Besides testing a correct run (complex trace) several Besides testing a correct run (complex trace), several errors have been introduced (missing wake-up, inject error in low power scheduler, wrong low power state of p g p MC, specification error).

23

DSN: Formal Conformance Testing of Power Traces H d M d l Hardware Model Trace Automaton

24

DSN: Formal Conformance Testing of Power Traces Software Model Verification Results

25

Conclusion

• Deployment Support Network (DSN)

– integrated into continuous integration tool chain including regression test – enables design, early deployment and testing of distributed embedded systems

• code distribution
• distributed unit test
• application control
• application control
• distributed trace analysis
• physical parameter testing

26

Overview

• Sensor Networks – A Critical Review
• Predictability and Efficiency

– Testing and Verification – Protocols – Energy Scavenging

• PermaSense

27

Motivation – Wireless fire-alarm system

• Wiring a building is cumbersome/costly or not possible
• Project collaboration with

Project collaboration with

– Large buildings requires scalable (multi-hop) solution

28

Requirements for a Fire-Alarm System

• 1. Event reporting1:

Al d l 10 d ( d t i k) – Alarm delay: ≤ 10 seconds (node to sink) – Robust and reliable

• 2. Status monitoring1:

– Node failure report: ≤ 5 minutes (at the sink)

• 3. Energy efficiency:

gy y

– Lifetime greater than 3 years (on 2xAA): < 1% duty cycle

→ Large deployments require multi hop solution → Large deployments require multi-hop solution

1 European Norm (EN) 54-25:2008-06, June 2008

29

European Norm (EN) 54 25:2008 06, June 2008

Link Analysis – Distance vs. Packet Reception Rate (PRR)

Gray Area

30

Link Analysis – Time vs. PRR (Link Stability)

88 % 88 % 12 % 12 %

31

Dwarf Algorithm: Network Structure

• Nodes are organized in rings

– according to their distance to the nearest sink

• The neighbors of node u are divided into parents Nu

−, peers

Nu

0, and children Nu +

– depending on the ring they belong to

Towards sink

32

Dwarf Algorithm: Node Observation

• Exchange status messages

A) Detect node and link failures B) K k h d l d

May interfere with

B) Keep wake-up schedules up to date

• Ring based status information aggregation

y alarm forwarding Ring based status information aggregation

– From the outer rings to towards the sink

33

Dwarf Algorithm: Alarm Forwarding

• Idea: Constrained flooding in combination with a delay-

Idea: Constrained flooding in combination with a delay aware nodes selection strategy

• Alarm is forwarded to k neighbors

– Degree k: redundancy vs message complexity Degree k: redundancy vs. message complexity

S l t f di d di th i k ti d

• Select forwarding nodes according their wake-up times and

relative positions

– Reduce alarm notification time Reduce alarm notification time

34

Dwarf Algorithm: Performance Evaluation

• 80 nodes based on real wired

placement

– provided by Siemens – 1-5 hops (3 on avg.) 2 fl – 2 floors

Verified by GloMoSim simulation

Sink

• Verified by GloMoSim simulation

P t t d

Sink

• Patented
• Part of future product

35

Conclusion

• Problem: Safety critical alarm reporting requires

– Robustness and timeliness and energy efficiency and gy y monitoring

• Protocol Solution: Dwarf

– Constrained flooding with delay-aware node selection – Ring based monitoring g g

• Integrated with reliable link estimation and initialization.

36

Overview

• Sensor Networks – A Critical Review
• Predictability and Efficiency

– Testing and Verification – Protocols – Energy Scavenging

• PermaSense

37

Motivation

[Heliomote: Srivastava] [Prometheus: Culler]

BTnode

C

What is different?

Conventional energy management:

How do we save energy ?

Energy harvesting:

In addition: When do we use energy ? In addition: When do we use energy ?

[Sunergy: June 2006]

System Model

Overview

System Model

Optimization problem: finite horizon control

t current time t t current state (memory, battery, …) current environment (input power)

System Model

Linear program (LP) specification (example)

Maximize minimal sensing rate

System Model

Optimization problem: finite horizon control

t

System Model

Optimization problem: finite horizon control

t

Solving a linear program (LP) in a resource-constraint sensor node resource-constraint sensor node at each time step ?

Multiparametric Linear Programming

Calculate optimal solution of the mp-LP

as an explicit function of the state vector p

Online complexity of mp-LP

sometimes acceptable sometimes acceptable …

O li l i f P Online complexity of mp-LP

9x7 matrices

sometimes not.

Bottleneck !!!

9x7 matrices

sometimes not. Bottleneck !!!

Hierarchical Control Design

Benefits

The upper layer avoids depletion of the energy

pp y p gy storage and increases robustness of the system

The complexity of the online controller is reduced

The complexity of the online controller is reduced significantly

Reduction: 83.0 % 91.0 %

Conclusion

Using methods from automatic control to achieve predictability and efficiency Cyberphysical Systems

Overview

• Sensor Networks – A Critical Review
• Predictability and Efficiency

– Testing and Verification – Protocols – Energy Scavenging

• PermaSense

51

PermaSense – Aims and Vision With PermaSense, we aim to:

– provide long-term high-quality sensing in harsh environments – obtain measurements that have previously been impossible (high resolution in time and space) impossible (high resolution in time and space) – provide relevant information for research or decision making

Reliability, delivery of information in near real- time, and integration of diverse sensors are , g ingredients for the next generation of early- warning systems.

52

PermaSense – The Matterhorn

• Located in Zermatt, CH
• 4478 m

4478 m

• Site of recent rockfall due

to extreme warming to extreme warming (07/2003)

– 25 nodes – Mutliple sensors – −40 to +65° C R kf ll d i – Rockfall, snow and ice, avalanches – 30 min. duty-cycle y y – 3 years unattended lifetime

53

PermaSense Site – The Deployment Region Hö li Rid Hörnli Ridge 3500 m asl

54

PermaSense – Transport

55

PermaSense – System Architecture

56

PermaSense – Sensor Stations on the Mountain

57

PermaSense – Base Station

• Powerful embedded Linux
• 4 GB storage, all data duplicated

g , p

• Solar power (2x 90W, 100 Ah, ~3 weeks)
• Backup modem
• Backup modem

58

Next Challenges

• Seismic/resistivity tomography

actuation and i sensing

59

Further Reading

60

http://www.btnode.ethz.ch

P di t bilit d ffi i Predictability and efficiency

influences the whole life cycle, and is integrated into all abstraction layers and is integrated into all abstraction layers.

Acknowlegement

• People:

– Jan Beutel – Federico Ferrari, Matthias Keller, Roman Lim, Andreas Meier, Clemens Moser, Mustafa Yuecel, Nikolay Stoimenov, Matthias Woehrle Woehrle

Funding:

• Funding:

– SNF Switzerland (via the NCCR MICS program) – BAFU BAFU – Competence Centre Environment and Sustainability (CCES) – Siemens Building Technologies – CTI: The Swiss Innovation Promotion Agency

62