Transcript Josh Bers BBN, Matt Welsh, Harvard

CitySense:
An Open, Urban-Scale Sensor Network Testbed
Josh Bers
BBN Technologies
Mobile Networking Systems Group
Matt Welsh
Harvard University
Division of Engineering and Applied Sciences
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
1
Sensor Network Testbeds
Goal: Support experimentation with wireless sensor networks at scale
●
●
●
Simulations are valuable but inherently limited
Understanding characteristics of real sensor networks in diverse environments
requires real testbeds and real applications
Testbeds should be open and easily shared by multiple research groups
CitySense: Planned outdoor testbed of 100 embedded PCs in
Cambridge, MA
●
●
●
●
Linux-based embedded PCs with meteorological and air quality sensors
802.11a/b/g interface with multihop wireless networking backbone
Collaboration between BBN Technologies and Harvard University
Funded by NSF under Computing Research Infrastructure program, 2006-2010
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
2
CitySense
Joint effort between BBN Technologies and Harvard University (Prof.
Matt Welsh, Co-PI)
●
●
●
NSF Computing Research Infrastructure (CRI) program grant (4 years), Rita Rodriguez
NSF Program Director.
BBN taking lead on hardware design and deployment planning
Harvard taking lead on software design and resource management
Goal: Deploy an outdoor, open wireless sensor network testbed
across the city of Cambridge, MA
Nodes consist of Linux-based embedded PCs with 802.11a/b/g
●
●
Mounted on top of light poles with assistance from City of Cambridge
Professional meterological sensor package for environmental monitoring
Web-based interface for job scheduling, debugging, profiling
●
●
Draw on experiences with MoteLab and extend to outdoor testbed
Open resource for the sensor network community
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
3
CitySense Overview
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
4
CitySense Overview
Vaisala
Mouting
mast
Fixture
Arm
Photocell (Power)
CitySense Node goes here
Power
input
Mounting
Straps
●
●
●
●
●
●
Vaisala meterological
sensor
WiFi
Antennas
Metrix embedded PC (Soekris single-board PC)
Runs Pebble Linux distribution
133 Mhz AMD processor
64 MB RAM and flash, 1 GB USB flash drive
Dual 802.11 a/b/g radios
Multiple sensors possible: weather, air quality, bio/chem
agents, webcams, microphones…
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
5
BBN Network Topology
• 3 Indoor nodes
plus gateway
• 2 nodes on roof of
buildings
– Racing
– Rosario
• Fully connected
except for
Gateway
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
6
Sensor Node Design Iter#1: Racing
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
7
Why CitySense?
Expand sensor networking testbeds beyond indoor deployments with
resource-constrained nodes
●
●
●
Outdoor testbed with large coverage area
Powered nodes with substantial CPU/memory/radio bandwidth
Provide blueprint for future sensor network designs and deployments
Shared resource open to research community
●
Leverage experience with Harvard’s MoteLab to provide shared experimental facility
Provide bridge to broader scientific communities
●
●
Partnership with Harvard School of Public Health – urban air pollution study
Educational impact at graduate, undergraduate, and K-12 levels
Connection to NSF GENI initiative
●
●
Shared facility for experimenting with sensor networks in realistic outdoor environment
Opportunity for connection to evolving network standards and support for
“Internet scale sensor networking”
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
8
CitySense sensor package
Vaisala Weather Transmitter WXT510
●
●
●
●
●
Wind speed and direction
Precipitation
Barometric pressure
Temperature
Relative humidity
Well-calibrated sensors, robust packaging for
outdoor environments
●
●
Designed for precise measurement of environmental
conditions
More accurate than typical component sensors used on motes
Serial interface for configuration and data access
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
9
Example data
Raw sensor ouput as received by our gateway via UDP packets multi-hopped from the sensor nodes:
●
●
●
●
Rain accumulation
Wind Speed and Direction
Pressure Temperature and Humidity
Sensor Status Data
Sensor data: net.citysense.sensors.PTHSensorOutput@1decdec Device-type=VAISALA WXT510
Device-name=0 Timestamp=Mon Mar 26 22:15:10 EDT 2007 Sample Interval=-1 Query command=N/A
Measurement airPressure value=1016.3 unit=hPa
Measurement airTemperature value=6.3 unit=Celsius
Measurement relativeHumidity value=89.5 unit=PERCENT
Sensor data: net.citysense.sensors.WindSensorOutput@12a54f9 Device-type=VAISALA WXT510
Device-name=0 Timestamp=Mon Mar 26 22:15:14 EDT 2007 Sample Interval=-1 Query command=N/A
Measurement directionAvg value=294 unit=DEGREES
Measurement directionMax value=330 unit=DEGREES
Measurement directionMin value=278 unit=DEGREES
Measurement speedAvg value=0.9 unit=METERS_PER_SECOND
Measurement speedMax value=1.2 unit=METERS_PER_SECOND
Measurement speedMin value=0.6 unit=METERS_PER_SECOND
Go to: http://citysense.bbn.com/ReadVaisala.pl for live data feed.
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
10
CitySense Networking
Most CitySense nodes will not have wired network connectivity
●
●
Several nodes (at BBN and Harvard) will act as gateways to the Internet.
Must use wireless mutihop network for all communications to nodes:
control/management, debugging, application traffic
Plan: Use multihop routing network based on OLSR
●
●
●
100's of meters range between nodes possible with appropriate antennas
Provide stable communications backplane with IP routing to individual nodes
User applications may implement their own routing protocols directly on 802.11 MAC
CitySense testbed will be timeshared across multiple users
●
●
●
CPU, memory, and radio bandwidth must be shared across applications
While not as limited as motes, this still raises some important resource
management questions
We expect demands on CitySense to vary widely across research groups.
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
11
CitySense Plug-and-Play Sensors
●
Sensor
Description
Document
maintains
●
On-node software enables easy
addition of new sensors
Adaptation layer defines a
common meta-data for sensors
to declare themselves to the
shared infrastructure
Sensor Adaptation Layer (SAL)
Device
Independent
Control API
Vendor-specific
sensor API
●
Sensor Adaptor
Meta-data are used to allocate
nodes to applications based
upon their sensing requirements
Sensor
Sensor
Sensor
Sensor
Hardware
Hardware
Hardware
Hardware
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
12
Open Challenges
Remote maintenance and programming
●
●
●
Physical access to nodes difficult or impossible
Must ensure software can be updated safely
Rollback to known-good “safe mode” if node loses network connectivity
Resource management and sandboxing
●
●
●
CitySense will be open to research community
How to prevent naïve or malicious users from dominating resources?
What are appropriate scheduling policies?
Application programming model
●
●
Should we allow arbitrary Linux binaries? Or require users to conform to constrained interface?
What distributed services should the system provide to applications?
Experimental support
●
●
Time synchronization, GPS vs. NTP
Distributed control: separate channel for management plane vs. in band
Some non-goals of this project…
●
●
Reinvent mesh networking: try to leverage existing solutions
Provide public Internet access: too latency sensitive; not appropriate for multihop mesh
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
13
GENI Wireless Research Enabled
●
Characterize URBAN RF environment: good urban propagation
models do not exist
●
Wireless Network Management
●
Dynamic RF channel selection
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
14
Summary
CitySense presents huge opportunity
for the sensor network community
●
●
Develop, deploy, and experiment with sensor networks at scale
in complex real-world outdoor urban environment
Shared research facilities for supporting diverse research groups
Planned 100-node outdoor testbed in Cambridge, MA
●
●
Linux-based embedded PCs with 802.11 and professional weather sensor
Planned future sensors include pollution/smog sensors.
For more information:
●
●
Josh Bers ([email protected]) and Matt Welsh ([email protected])
http://www.citysense.net
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
15
Related Work / Facilities
WINLab, ORBIT Rutgers [Raychaudhari ]
ENL, USC motes [Govindan]
sMote, Berkeley [Culler]
RoofNET, MIT [Morris, et. al]
U Colorado [Sicker & Grunwald]
Others…
Community networks:
CUWin, Corpus Christie, TEX, etc.
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
16
Acknowledgements
BBN
●
●
Abhimanyu Gosain, Tufts Intern
Frank Bronzo
Harvard
●
●
●
●
●
●
Amal Fahad
Jon Hyman
Kevin Bombino
Geoff Mainland
Rohan Murty
Matt Tierney
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
17
Current Status
BBN Testbed
●
●
3 indoor nodes
2 outdoors with weather sensors
Node Design
●
●
2 Prototype designs tested
Working on City approval of streetlight mounted enclosure
Wireless Network
●
●
OLSR mesh active
Characterized basic performance
City Streetlight Mounting
●
Received approval from City of Cambridge
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
18
Next Steps
BBN & Harvard Testbeds
●
●
Grow size of each testbed to ~ 10 nodes outdoors
Link 2 networks via advantaged nodes
Wireless Network
●
●
Characterization:
●
Establish performance benchmark suite
Management plane:
●
Test high-power, 700 mW, 900 MHz radios (ubiquiti
networks)
City Deployment
●
First Nodes targeted for Summer-Fall ‘07
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
19
Preliminary Results: Urban RF Activity
From BBN’s rooftop mounted nodes
●
Total 5MHz Channels in use: 29 out of 74
●
●
●
802.11b/g: 11/14
802.11a: lower 11/40, upper 7/20
Total devices seen (distinct MAC addresses)
●
●
in 15 days: 205
in 12 hours: 25
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
20
Collaborators / Target Users
Magid Ezzati: Co-PI Harvard School of Public Health  Urban
pollution studies
Ken Mandl: Director of CHIP’s program Childrens Hospital, Boston 
real-time tracking of ER symptom reports
David Gute: Tufts University EE department: water quality sensors
Tom Little: BU EECS: video sensors
Chris Rogers & Marina Bers: Tufts EE: Educational Outreach  K-12
curriculum in sensor nets.
© 2007 Matt Welsh – Harvard University
© 2007 BBN Technologies
21