150 likes | 206 Vues
Towards Runtime Support for Energy Awareness in Wireless Sensor Networks. Thomas Trathnigg and Reinhold Weiss Institute for Technical Informatics Graz University of Technology Graz, A-8010 Austria {trathnigg,rweiss}@iti.tugraz.at. Outline. Introduction Measuring Energy in WSNs
E N D
Towards Runtime Support for Energy Awareness in Wireless Sensor Networks Thomas Trathnigg and Reinhold Weiss Institute for Technical Informatics Graz University of Technology Graz, A-8010 Austria {trathnigg,rweiss}@iti.tugraz.at
Outline • Introduction • Measuring Energy in WSNs • Measurement Setup • Error Analysis • Validate PowerTOSSIM • Conclusion + Outlook
Introduction • Lifetime of a wireless sensor network depends on the energy consumption of each node • Limited energy-budget • Battery-powered • Energy harvesting • Energy-awareness • Energy-aware routing • Dynamic Power Management • …
Introduction • Monitor the energy consumption of each mote • Online monitoring • Simulator calibration • Requirements • accurate • low-power • small • inexpensive
Mica2 Motes • ATMEGA 128L • 7.3 Mhz 8-bit CPU • 128 KB code, 4 KB RAM • 433, 868 or 916 Mhz, 76.8 Kbps FSK radio transceiver • 512 KB flash for logging • Sandwich-on sensor boards • Powered by 2 AA batteries
Typical Current Profile of Mica2 • 14-bit 100MHz dual-channel Digitizer (National Instruments) • Fast changes in current profile due to cpu and radio state changes • Clamp-on current probes • „Fuel Gauges“ • Based on peridodical ADC sampling • Energy-Driven Sampling • Based on an approach published by Chang et al. • Detecting software hotspots on a PDA
Error Analysis • Non-ideal behavior of electrical components • discharge time of capacitor • 500ns, 1000:1 ratio at 35mA; 0.1% error • Voltage at the mote • voltage drop caused by shunt resistor • <2% error • Current-sense amplifier • error is below 2% in the range 3 to 66mA • Bandwith of current-sense amplifier • fastest current change measured on mica2 mote 2.4mV/s • We expect the error of our setup to be below 5%
Calibration + Verfication • Determined the amount of energy a ramp depicts (54J) • Verification • Stable voltage supply 3V • Constant current load • 60s measurments • Comparision of measurement with calculated result
Validate PowerTOSSIM • TinyOS 1.1.15 • PowerTOSSIM • Support only for mica2 • used CPU cycle counting • mica2 (868MHz) • Deluge disabled • Several TinyOS Applications measured for 60s
Analysis of PowerTOSSIM Divergence • Possible reasons • Measurement errors • Inaccuracies in the simulation PowerTOSSIM simulates at 4MHz, mica2 motes operate at 7.38MHz • Power-model of PowerTOSSIM systematic errors values of the power-model may be inaccurate 443MHz vs 868MHz Other hardware differences
Energy Consumption of Different Motes • No mica2 motes at 443MHz available, so we checked only other hardware differences • 8 different mica2 motes • 4% max. difference for Blink • 3.4% max. difference for CntToRfm
Conclusion + Outlook • Approach of energy-based sampling is feasible • size • cost • low-power, should be improved • accuracy • range must be increased • Improve/calibrate energy model of PowerTOSSIM • Redesign of measurement setup • measurement range • low-power • integration on mica2 sensorboard