GDI Data Analysis

1
GDI Data Analysis

• Robert Szewczyk
• December 20, 2002

2
Global statistics

• 43 distinct nodes reporting data at various times
• 1132548 packets logged in the DB
• 3 maintenance events, roughly every month
• Heavy mote losses 4 daily
• Best nodes nearly 90000 packets on a pair of AA
  batteries, over 2.5 months unattended operation

3
Sample data collected

• Collected light levels, temperature, relative
  humidity, thermopile IR and ambient temperature

4
Sample data sensor discussion

• Thermopile sensor -- difficult to gain the
  confidence in the readings
• Good correlation between the ambient temperature
  readout from the thermopile and the measured
  temperature
• But difficult to scale it
• Difficult to determine appropriate signal
  processing for the IR signal
• Low confidence even as a occupancy detector
• Pressure sensor
• Not used in the application, produced even less
  reliable nodes
• Humidity sensor
• packaging problems bulky, battery drain when
  wet
• addressed in the next revision of the weather
  board
• Temperature sensor
• resolution significantly lower than advertised
• reported in a plausible temperatures in a wide
  range (-10 60 deg C)
• Very good correspondence with Coast Guard data
• Light sensor
• Known limitations, good baseline indication of
  mote health

5
Power Management

• Expected 6 months _at_ 3 duty cycle
• Real world performance MUCH worse best node
  lasted only 2.5 months
• Correlation between packet success rate and
  battery voltage
• Boost converter provides less consistency than
  expected
• Batteries can be drained down to 0.8 V per cell,
  poor reliability below 1.1 V per cell

Battery voltage at node 57, the most reliable
mote from the initial deployment. Last packet
from that node on that set of batteries was
received on 9/24 the node reliability declined
drastically after 9/22.
6
Network analysis

• Several underlying causes for packet loss
• Laptop / database crash connection to the
  laptop was only available 47 of the time
• Low battery levels
• Collisions
• Environmental conditions wind blowing antennas
  out of alignment, rain affecting humidity sensor
  and short-circuiting the battery
• Packet loss distribution
• Packet loss does not behave like an independent
  distribution
• Work in progress to bin the potential causes of
  packet loss

7
Loss distribution
8
(No Transcript)
9
Phase stability
10
Conclusions

• First application that stressed low power and
  unattended long-term operation
• What we learned
• Need a lot of diagnostic information to support
  long running apps
• Gain confidence in the sensor readings
• Diagnose and remotely repair faults (if
  possible) provide bounded downtime
• Components addressing many concerns either exist
  or will soon exists link layer acks, channel
  monitoring component, watchdog timer, etc.
• Boost converter falls short of expectations
• Poor efficiency
• Reduced performance on weaker batteries

11
Conclusions (cont.)

• Future work
• Application redeployment in a more controlled
  environment
• Further root cause analysis
• Incorporating the lessons learned into Generic
  Sensor Kit and second generation weather board
• Accessing GDI data
• http//www.greatduckisland.net
• PostgreSQL database
• Server dbsvr.berkeley.intel-research.net
• Username reader
• Password readonly
• Database gdi
• Most interesting table weather

12
Mote 18 Outside
13
Mote 26 Burrow 115a
14
Mote 53 Burrow 115b
15
Mote 47 Burrow 88a
16
Mote 40 Burrow 88b
17
Mote 39 Burrow 84