Deborah Estrin, Ph.D.

1
Embedding the InternetHow Smart Sensors May Help
Save the Planet

• Deborah Estrin, Ph.D.
• http//cens.ucla.edu/Estrin destrin_at_cs.ucla.edu
• Work summarized here is that of students, staff,
  and faculty at CENS
• We gratefully acknowledge the support of our
  sponsors, including the National Science
  Foundation, Intel Corporation, Sun Inc., Crossbow
  Inc., Agilent, Microsoft Research,, and the
  participating campuses.

2
Why wireless sensor networks?

• Many critical issues facing science, government,
  and the public call for high fidelity and real
  time observations of the physical world
• Networks of smart, wireless sensors will reveal
  the previously unobservable
• Designing physically-coupled, robust, scalable,
  distributed-systems is challenging
• The technology will also transform the business
  enterprise, from the factory floor to the
  distribution channel

3
Embedded networked sensing will reveal previously
unobservable phenomena

• Remote sensing transformed observations of
  large scale phenomena
• In situ sensing transforms observations of
  spatially variable processes in heterogeneous and
  obstructed environments

Red Soil Green Vegetation Blue Snow
SPOT Vegetation Daily Global Coverage SWIR 3 Day
Composite
Predicting Soil Erosion Potential Weekly MODIS
Data
Sheely Farm 2002 Crop map
San Joaquin River Basin Courtesy of Susan
Ustin-Center for Spatial Technologies and Remote
Sensing
4
Environmental monitoring applications exhibit
high spatial variations and heterogeneity
Precision Agriculture, Water quality management
Overflow of embankment
Algal growth as a result of eutrophication
Impact of fragmentation on species diversity
5
Approach

• Embed numerous, low-cost, distributed devices to
  monitor and interact with physical world
• Deploy spatially and temporally dense, in situ,
  sensing and actuation

• Network these devices so that they can
  coordinate to perform higher-level identification
  and tasks
• Requires robust distributed systems of thousands
  of devices.

6
The network is the sensor!

• Requires large distributed systems with adaptive
  internal behavior that can report spatio-temporal
  events, and characterize phenomena, not just
  return individual temporal and spatial data
  points.
• Model based anomaly detection drives additional
  data/sample collection, field observation

Model basedanomalydetection
7
Technical challenges

• Physical environment is dynamic and unpredictable
  
• Small wireless nodes have stringent energy,
  storage, communication constraints

• In-network processing of data close to sensor
  source provides
• Scalability for densely deployed sensors
• Low-latency for in situ triggering and adaptation
• Embedded nodes collaborate to report interesting
  spatio-temporal events

8
A participants (biased/limited) view of history
Early history Ubiquitous computing/Smart
Spaces/Pervasive ORL (UK), Xerox PARC, MIT Media
lab, IBM, HP, DARPA DSN (Kahn, Cohen (USC/ISI),
) DARPA Packet Radio program
Almost a decade of wireless sensor networks
research programs DARPA WINS and AWARES 96-98
UCLA Kaiser-Pottie, IETF MANET WG (ad hoc
routing), 802.11 WG ISAT Simple Systems study
98 Estrin, Pottie, Weiser, Clark, Paxson,
ISAT Robotic Ecology Pottie DARPA SenseIT
99 USC/ISI, Cornell, Xerox, UCB, BBN, Penn
State, Univ Ill., MIT NRC Embedded Everywhere
00 Smart Dust and TOS papers 00 Intel Berkeley
lablet, Startups Crossbow, Ember, Dust,
Sensicast DARPA NEST 01 UCB TinyOS, Ohio
State, Univ Virginia, MIT, Intel/Xbow NSF ITRs,
CENS STC(UCLA-USC), CITRIS (UCB) NSF Sensors
and sensor networks NETS/NOSS Industrial RD
MSR, Nokia, IBM, HP, PARC, Motorolla, Sun,
Agilent, Intel
Sigcomm
SOSP/OSDI
IPSN 2002
Sensys 2003
WSNA
Mobicom
Mobihoc
Emnets
DCOSS
ICASP
9
Decade of Networked Sensor Node Developments
LWIM III UCLA, 1996 Geophone, RFM radio, PIC,
star network
AWAIRS I UCLA/RSC 1998 Geophone, DS/SS Radio,
strongARM, Multi-hop networks
Sensor Mote UCB, 2000 RFM radio, Atmel
Telos Mote UCB, 2004 Zigbee radio, Motorolla
10
TinyOS/Mote - Open Platform (Adapted from Culler)

• De facto std in sensor nets with active open
  source community
• www.tinyos.net, tinyos.sourceforge.net
• several platforms rene, mica, dot, mica2,
  bosch, iMOTE, dust, micaZ, Telos

WeC 99 Smart Rock
Small microcontroller - 8 kb code, - 512
B data Simple, low-power radio - 10
kb EEPROM (32 KB) Simple sensors
Crossbow
11
TinyOS (Adapted from Culler)

• Framework for app specific OS
• communication centric, event driven, modular
• Expressed in nesC permits whole-system
  optimization
• Setting for CS explorations across the stack

Applications Compose Just What they need
Tracking Application
Sensing Application
Multiple Network Layer Protocols
12
SOS Operating System(R. Shea, S. Han, M.
Srivastava, E. Kohler, et al)

• Sensor networks require uninterrupted operation
  despite needed post-deployment software updates
• SOS supports dynamic insertion of binary modules
  onto the running kernel as low energy solution
• Module distribution is less expensive than full
  binary distribution / patching used in static
  operating systems
• Binary execution is more efficient than
  interpreted code execution used on virtual
  machines
• SOS modular programming and execution facilitates
  sharing of underlying modules by different
  applications running on a single sensor network

13
Sensor DiversityEmbedded mote-based imaging
(Cycl o ps)

• Inference in optical domain
• CMOS technology Low power ( capture lt 40mA)
• Cyclops is not imager but rather a sensor
• Small picture size Target below 256256
• Example Applications
• Color estimation Monitor triggering,
  Agriculture, Motion detection, Security
• Low power, long term image archival phonology
• Platform
• Atmega128 8bit RISC PROCESSOR
• 512 KByte of Flash for local File system
• 512 KByte RAM Enough room for heavier computation
• Software and algorithm innovations
• in-network processing of images for event
  detection
• Limited resources, but in limited context

Mohammad Rahimi
14
Preliminary Cyclops power consumption measurements

• Hardware
• Low power by on demand access to resources
• Simple interface
• Embedded Software
• An inference sensor
• General Purpose libraries for image manipulation

First Generation of Cyclops
15
Heterogeneity is key to deployed systems and
the field as a whole

• Several classes of systems
• Mote herds Scale
• Collaborative processing arrays Sampling rate
• Networked Info-Mechanical Systems Autonomy
• Achieve longevity/autonomy, scalability,
  functionality with
• heterogeneous systems
• in-network processing, triggering, actuation

lifetime/autonomy
Mote Clusters
Infrastructure- based mobility(NIMS)
scale
Collaborative processing arrays (imaging,
acoustics)
sampling rate
16
Tiered Architecture(Joint work with Eddie
Kohler, Ramesh Govindan, et al)

• Currently much of sensor networks research
  accepts following architectural principle
• We believe it is reasonable to assume that
  sensor networks can be tailored to the sensing
  task at hand. In particular, this means that
  intermediate nodes can perform application-specifi
  c data aggregation and caching
• D. Estrin, R. Govindan, J. Heidemann, S. Kumar,,
• Next Century Challenges Scalable
  Data-Dissemination in Wireless Sensor Networks,
• Proc ACM Mobicom 1999.
• However, application-specific programming of
  mote-class devices is fundamentally hard due to
  many design constraints Processor, Memory,
  Energy, Sensor noise, Wireless

17
Tiered Embedded Networks (Tenets)(Joint work
with Eddie Kohler, Ramesh Govindan, et al)
Internet

• Optimize large node (master) for multi-node
  data fusion functionality and complex application
  logic
• Optimize small node (mote) for broadest
  distribution

Local Data
Masters Powered (currently) devices, wireless
mesh for communications
Motes Relatively impoverished devices, small
depth clusters
Adapted from R. Govindan
18
Rationale

• Masters relatively unconstrained
• Natural aggregation, correlation, processing
  point for mote field data
• Low diameter mote cloud desirable (3-4 hops)
• Deeper networks exhibit poor performance given
  wireless packet losses in low power radios (and
  PR Kumar scaling limits)
• But multi-hop is fundamental for coverage,
  flexible deployment
• Larger networks built out of collections of
  masters and associated mote clouds
• In situ, in network coordination and processing
  needed across masters for latency and scalability

19
The Tenet Architecture
(Distributed) computing substrate Application
development happens here

• Simplify this interface!
• sensor-addressable tasking
• (reliable) data collection
• generic signal processing
• this can evolve

Programmable data acquisition
Local task processing, event detection, duty
cycling, MAC, localization, time synch, tree
routing, congestion control etc. here
20
Need New S/W environments for 32-bit nodes

• Logistical and environmental issues in deployment
• Fielded systems tend to degrade more quickly than
  in the lab
• Environmental conditions weather, animals, RF
  and sensor channel
• Uniform deployments are difficult to achieve
  node replacement
• Observed Data can cause unexpected failures, new
  bugs in the field
• e.g. Acoustic ranging system encountered new
  kinds of noise, leading to new kinds of
  inconsistencies in geometry, crashing Non-Linear
  Least-Squares (NLLS) algorithm but not
  reproducible in the lab

?
21
EmStar development environment

• EmStar is a layer above Linux designed to enable
• Robustness Keep system running despite
  unexpected failures and bugs
• Visibility Easily debug/diagnose running systems
• Simulation, Emulation Rapid iteration via
  real-code simulation tools
• Module Reuse Leverage existing libraries, tools,
  and services
• EmTos Wrapper library provides TinyOS API and
  Services

Robust multi-process, microkernel architecture
Simulation Framework with real RF channels
Visualization Tools
22
Libraries, Tools, Services

• Libraries and IPC Support
• FUSD IPC via device file interfaces
• Device Patterns Libraries that provide standard
  kinds of devices.
• Status Device, Packet Device, Sensor Device
• EmTOS NesC/TinyOS compatibility wrapper
• Tools
• EmRun Manage running EmStar processes and
  collect logs
• EmSim/EmCee A real-code simulator that can
  support real radios
• EmView A visualization tool
• Services
• Link/Neighborhood estimation
• Time Synchronization
• Routing Flooding, Sink Tree, Diffusion

23
EmTOS Support for Heterogeneous Systems in Emstar

• Wrapper Library
• Provides TinyOS API and Services
• Enables NesC to provide new EmStar services
• Compiles NesC Application EmTOS library into a
  single EmStar module
• Benefits
• Simulate systems of motes and microservers in
  same world
• Easy porting of TinyOS/NesC services to
  microservers
• ESS2
• TinyDB

24
EmSim/EmCee

• EmSim a real code simulation environment for
  EmStar
• Runs N copies of an EmStar system on a single
  machine
• Each node gets its own device namespace
• Sim components provide interface to simulated
  world
• sim_radio models an RF channel and MAC layer
• sim_sensor models or replays sensor data
• EmCee simulated nodes use real radios for comm
• Runs N copies of an EmStar system, connects each
  nodes link device to a real Mote radio
  connected by a serial multiplexer

25
Visibility and Debugging

• Why is Visibility important?
• Reveals internal state of modules
• Reveals traffic between modules, e.g.
• Observe when each neighbor update is issued
• Observe data traffic through network stack
• How Browsable Device File Hierarchy
• Similar to /proc, modules report their internal
  state
• Human readable and binary versions
• Binary channel used for IPC
• Same info visible interactively from the shell
• Enables Debugging
• Locate faults by verifying modules input and
  output
• Visualize distributed system including dynamics
• In simulation
• In real life with debugging backchannel

cat node001/link/mote0/status Root Device
Simulated BMAC Mote Stack sim,mote0 Interfac
e Addr 0.0.0.1 MTU 200 Stats packets_rx 1
packets_tx 1 bytes_rx 164 bytes_tx 164
errors_tx 0 errors_rx 0 Active 1 Promisc
0 POT 6
26
Robustness and Fault Tolerance

• Why is robustness so important?
• Degradation in presence of permanent HW, SW
  faults
• Recovery from transient faults, limiting
  cascading failures
• e.g. unanticipated sensor data
• Unusual cases that yield inconsistent or
  confusing data
• Microkernel Implementation
• FUSD (Framework for User Space Devices)
• Fault isolation between client, server, and
  kernel
• Servers robust to faulty clients
• Modules communicate through POSIX Device API
• Inter-module Fault Tolerance
• Similar approach as in distributed systems
• Survive a range of errors and module failures
• Soft-state protocols between modules
• Rate limiting, filtering, refresh at module
  interfaces

27
Status Many first generation solutions existin
TinyOS and Emstar based tiered systems
Localization Time Synchronization
Self-Test
In Network Processing
Programming Model
Routing and Transport
Event Detection

• Reusable, Modular, Flexible, Well-characterized
  Services/Tools
• Routing and Reliable transport
• Time synchronization, Localization, Self-Test,
  Energy Harvesting
• In Network Processing Triggering, Tasking, Fault
  detection, Sample Collection
• Programming abstractions, tools
• Development, simulation, testing, debugging

28
Environmental Application Drivers at CENS

• Contaminant Transport, Soils
• Three dimensional soil monitoring
• Error resiliency at node and system level
• Data assimilation, model development
• Marine microorganisms
• Aquatic operation
• Micro-organism identification
• Sensor driven biological sample collection
• Biology/Ecosystem Processes
• Robust, extensible microclimate monitoring
• Image and acoustic sensing
• Infrastructure based mobility

29
Wastewater reuse in the Mojave Desert
Reclaimed wastewater irrigation pivot plots

• Where does the County Sanitation District (CSD)
  of Los Angeles put 4 million gallons per day of
  treated wastewater in a landlocked region?

Palmdale, CA wastewater treatment plant
Nitrate sensor mimicking plant root fibers
30
Plankton dynamics in marine environments
Spatial and temporal distributions of harmful
alga blooms (red, green, brown tides) in marine
coastal ecosystems
Experimental and observational studies of
chemical, physical and biolgical features
promoting bloom events
31
EmPack focus on usability, tools
32
Current technology research focus
Objectives
Constraints

• Embeddable, low-cost sensor devices
• Robust, portable, self configuring systems
• Data integrity, system dependability
• Programmable, adaptive systems
• Multiscale data fusion, interactive access

• Energy
• Scale, dynamics
• Autonomous disconnected operation
• Sensing channel uncertainty
• Complexity of distributed systems

33

Real time fusion enablesinteractive data access
and visualization in the field

• Emissary
• Real time access to archived data and data
  models from the field
• Contextualize in situ observations
• Guide data collection, system debugging

transform physical observations from batch to
interactive process
34
Next generation heterogeneous systems
includemobility, sensor diversity, fusion,
multi-scale

• Infrastructure assisted mobility and
  actuation offer
• Sensor diversity
• Location
• Type
• Duration
• Magnified effective sensor range
• Dense, adaptive, fidelity-driven, 3-D sensing
  and sample collection
• Challenging algorithms

Figure courtesy of Bill Kaiser
35
From data to observationsfusing data from
multiple scales in real time

• Satellite, airborne remote sensing data sets at
  regular time intervals
• Coupled to regional-scale backbone sensor
  network for ground-based observations
• Need fusion, interpolation tools based on
  large-scale computational models

Example identification of invasive riparian
species using HyMap (airborne hyperspectral
scanning)
images from Susan Ustin, UC Davis
36
Data integrity in sensor networks multilevel
calibration aided by in situ, interactive access

• Bench-top calibration
• Pilot deployment
• develop in situ calibration protocol
• characterize longevity, degradation
• Early in the deployment
• Take advantage of the sensors integrity
• Calibrate model (distributed parameters)
• Integrate DAQ with simulator to accelerate
  process
• Later (as sensors become suspect)
• Reverse the process
• Let the network identify bad sensors Self-Test
• Incorporate uncertainty into the process

37
Science applications are historical drivers for
information technology development and deployment

• Early embedded sensing applications
• Biological and Earth Sciences
• Environmental, Civil, Bio Engineering
• Public health, Medical research
• Agriculture, Resource management

• Science is early adopter because the technology
  is transformative and research tolerates risk
• The same technology will transform the business
  enterprise
• Important historical precedents
• Weather modeling--early computing
• Scientific collaboration--Internet
• Experimental physics (CERN)--WWW
• Computational science--Grid computing

• Embeddable device developments
• Energy-conserving platforms, radios
• Miniaturized, autonomous, sensors
• Standardized software interfaces
• Self-configuration algorithms

38
CLEANER will be an integrated network to support
fundamental engineering research and education
on large-scale, environmental problems. It will
provide researchers across the nation with access
to leading-edge linked sensing networks, data
repositories, characterization and computational
tools for integrated assessment
modeling,connected through high performance
computing and telecommunications networks.
Modeling would be the central component for
analysis
NEON will transform ecological research by
enabling studies on major environmental
challenges at regional to continental scales.
Scientists and engineers will use NEON to conduct
real-time ecological studies spanning all levels
of biological organization and temporal and
geographical scales.
Biogeochemical cycles, Biodiversity, Climate
change, Invasive species, Infectious diseases,
Land use change, Hydrology
39
Engineering, civilian, enterprise
applicationswill eventually dominate

• As the technology matures we will find
  wide-reaching applications in the built
  environment and throughout the business
  enterprise.

40
Pervasive observation in the public sphere
Transparency Visibility
Privacy Reframed
Design versus Regulation
Courtesy of Dana Cuff - Institute for Pervasive
Computing and Society
41
Many technical and policy challenges ahead
How will we monitor the monitors?
Multi-scale data fusion
Embeddable sensors
Trustworthy, autonomous, distributed systems
42
Personal observations/lessons

• Take smaller steps (there are interesting and
  useful points along the way)
• Hundreds, not hundreds of thousands
• Semi-autonomous, not only fully autonomous (start
  by enhancing human capabilities)
• Use infrastructure to deploy initial systems and
  learn from them (node localization, mobility,
  cellular and wifi backbones)
• Tiered systems are fundamental to longevity,
  system effectiveness
• Avoid wireless scaling limitations (PR Kumar)
• Thoughtful distribution of in-network processing,
  coordination
• Build on existing system tools
• Provide data in context
• Start with systems whose in situ data augments,
  enhances existing data
• Integrate with models, other data sources
• Support interactive access to in situ data
• Top priority is to deploy, use, learn
• First generation systems/technology exist
• As technology matures, deployable systems will
  become increasingly powerful (modalities,
  precision, scale, cost)

43
Broad relevance to global issuesrequires
commitment to multidisciplinary experimental
research
Civil Infrastructure
Security
Global Climate Change
Precision Agriculture
Wildfire Management
Public Health
Coral Reef Health
Water Quality
Early Warning, Crisis Response
Global Seismic Grids/Facilities
Commercial development needs and opportunities
Data Management
EmbeddableSensors
Programming
Adaptive Sampling
Robotics
Tools
Embedded Imaging
High Integrity Systems
44
CENS Research Organization Road Map
45
For further investigation

• Center for Embedded Networked Sensing,
  http//cens.ucla.edu
• TInyOS and Mote platforms UC Berkeley, Intel,
  Crossbow, Sensicast, Dust Networks, Ember
  http//www.tinyos.net
• NSF Workshops including Sensors for Environmental
  Observatories, http//www.wtec.org/seo/seo6.htm
• National Ecological Observatory Network,
  http//www.neoninc.org
• Principles of Embedded Networked Systems Design,
  Gregory J. Pottie and William J. Kaiser,
  Cambridge University Press, Spring 2005

Contact Deborah Estrin destrin_at_cs.ucla.edu