Transcript Power evaluation of SmartDust remote sensors

Power evaluation of SmartDust remote
sensors
CS 252 Project Presentation
Robert Szewczyk
Andras Ferencz
Spring 2000, 4/27/00
Application: remote sensor
• Periodic measurements
– light, temperature, humidity
• Data processed in the infrastructure
– thin client model
– communication is necessary
• Participation in routing protocols
• Unattended mode of operation
Spring 2000, 4/27/00
Platform: SmartDust
• Low-power wireless communication
– RFM TR1000 transceiver, bit-level interface
• Range of digital and analog sensors
– Light sensor - photo resistor
– Temperature sensor - I2C interface
• Low-power microcontroller
– ATMEL AVR 90LS8535, Harvard architecture,
8KB program, 512 byte data
Spring 2000, 4/27/00
Mapping
• TinyOS framework
– software modules consisting of
• event handlers
• threads to perform arbitrary computation
asynchronously
– hardware abstraction or replacement
• RFM bit-level interface
• byte-level radio interface, similar to UART
– active message-like communication scheme
and execution model
• Crucial resource: energy
Spring 2000, 4/27/00
Initial evaluation
•
Methodology:
– logic analyzer timing diagrams
– processor power consumption from datasheets
– RFM power measurements
•
Wireless communication costs
–
–
–
–
–
2.0 μJ/bit radio cost
software costs, going from bits to bytes: 690 nJ/bit
longest path through the time-critical code: 40 μs
communicating processor at 4MHz idle 50% of the time
radio draws constant power regardless of data rates
Spring 2000, 4/27/00
Experimental setup
• Tools
– HP 16550A logic analyzer
– HP 16532A digital oscilloscope
– 2.84V DC power supply
• Current measurements
– 10 Ohm 5% tolerance in series with mote
– data point extraction from oscilloscope images
– typical settings: 1 ms total interval analyzed, dynamic
range: 160 mV
– differential analysis to extract contributions of individual
components
– typical variation of successive experiments: 5%
Spring 2000, 4/27/00
Measurements
AVR 90LS8535, 2.84 V@4MHz
Instruction
type
Energy per
cycle (nJ)
Energy per
instr (nJ)
Device
idle
1.70
1.70
LED
noop
3.39
3.39
arithmetic/
logic
3.41
memory
read*
memory
write*
Energy per CPU
cycle
Energy per
quantum
1.89
1.89
nJ/cycle
Photo
0.08 - 0.28
0.08 - 0.28
nJ/cycle
3.41
ADC
0.36 - 0.30
4.62 - 3.95
nJ/conversion
3.66
7.32
RMF send
100 μs pulse
2.56
2050
nJ/bit
3.75
7.50
RFM receive
2.44
1950
nJ/bit
*memory instructions take 2 cycles
Spring 2000, 4/27/00
Exploration
• Implications
– current configuration: data rates up to 25Kbps or can
reduce clock speed by a factor of 2
– dedicate a more sophisticated interface to the radio
– speed up the transmission rate: transmit and turn off
• Research question:
– should we dedicate a separate microcontroller to each
IO device?
– Evaluate 2 processor system:
• a processor dedicated to the radio
• a processor dedicated to other sensors
• UART communication between subsystems
• scale frequency and voltage to minimize power
usage
Spring 2000, 4/27/00
Methodology: Power aware simulator
• ATMEL AVR instruction-set simulator
– power-aware
• incorporate the measurements from the real system
– IO device simulation
• timers, pins, and UART
• use per cycle energy data from the real measurements
– thread safe (need to simulate a multiprocessor system)
• Communication system
– Initially a UART evaluation
– Shared memory models
• TinyOS application
– TinyOS naturally supports a multiprocessing environment
– split the application at the byte-level radio
Spring 2000, 4/27/00
Results and conclusions
• Simulator status
– tested single processor configuration, agreement with
empirical measurements
– dual processor configuration in progress
• Estimates
– RFM processor - run at 2MHz, 5% idle, require 3mA
current
– Master processor - can run as slow as 200kHz, in order
to handle peripherals
– Inter-processor communication costs related to
interconnect are small (cf. UART data)
– Inter-processor communication costs related to
software overhead are significant (interrupt handling,
busy waiting or inability to power-down)
Spring 2000, 4/27/00
Acknowledgments
• SmartDust members
– Kris Pister, Seth Hollar
• TinyOS (ASPLOS 2000 submission)
– David Culler, Jason Hill, Rob Szewczyk, Alec
Woo
Spring 2000, 4/27/00