Kassandra Charalampidou

Transcription

1 Fidelity and Yield in a Volcano Monitoring Sensor Network Geoff Werner-Allen, Konrad Lorincz, Jeff Johnson, Jonathan Lees and Matt Welsh OSDI 2006 October 19th, 2010 Duration: 20 min Kassandra Charalampidou Titel van de presentatie 1

2 Experiment 19-day network deployment Reventador volcano - Equador 16 sensors Seismic and acoustic data Data fidelity (Quality) Yield (Quantity) Fidelity and Yield in a Volcano Time difference Monitoring Titel van in Video Sensor social presentatie Streaming networks Network 2

3 Sensor Network Challenges High-resolution signal collection High-frequency sensing for volcano activity Triggered data collection It is impossible to continuously transmit high resolution signal Timing accuracy Signal must be stamped with high accuracy Fidelity and Yield in a Volcano Time difference Monitoring Titel van in Video Sensor social presentatie Streaming networks Network 3

4 Contents Network Topology Event detector accuracy Data collection performance Time Accuracy Conclusions Fidelity and Yield in a Volcano Monitoring Titel van de Sensor presentatie Network 4

5 Background Standalone dataloggers But: large and power-hungry Impossible deployment in remote or hazardous area Data retrieved manually Significant effort Fidelity and Yield in a Volcano Time difference Monitoring Titel van in Video Sensor social presentatie Streaming networks Network 5

6 Sensor Node Fidelity and Yield in a Volcano Monitoring Titel van de Sensor presentatie Network 6

7 Network Topology (1/2) Network designed for high-resolution signal collection Nodes are relaying data via a radio-modem to base station GPS receiver establishes a global time base Continuously sampling at 100 Hz 7

8 Network Topology (2/2) Storing data to flash memory of 1Mb 256-byte blocks 20 min of data Status messages every 10 sec Base station can issue commands 8

9 Data Collection (1/2) Event detection algorithm Report to base station (according to a threshold) Triggers from 30% of nodes within a 10 sec window Base station makes a request of the last 60 sec of data 9

10 Data Collection (2/2) Base node sends command to nodes to stop sampling Nodes send data as series of 8 packets Base node sends command to resume sampling During data transfer nodes disable sampling 10

11 Results 11

12 Node Downtime 3-day outage due to software bug Synchronization faults Base failures due to power outages 12

13 Event Detector Accuracy (1/2) Location of node Orientation quality of seismometer Not related to the distance from the vent 13

14 Event Detector Accuracy (2/2) 14

15 Data Yield (1/2) Depth to the base station Highest yield for nodes within two hops from root Radio link quality 15

16 Data Yield (2/2) 16

17 Fetch Latency Main factor: Depth of the node in the routing tree Great difference between nodes at 1 and 6 hops from sink 17

18 Time Synchronization (1/2) Flooding Time Synchronization Protocol (FTSP) GPS receiver records GPS time and corresponding FTSP time Sends this information to the base station 18

19 Time Synchronization (2/2) Each node records its local l time and the corresponding FTSP time. Every data block carries local time we can recover the exact time of each block Three timebases: local time, global time by FTSP, GPS time Map local to global Map global to GPS-based GMT. FTSP instability Timestamp rectification 19

20 Conclusions Many back-to-back events missed due to lack of simultaneous recording and transferring. Too many failures of base station, result in low system uptime. Although, no power-related failures experienced. Event detection parameters required tuning to capture more signals. Correcting the timing of signal was a big challenge. Fidelity and Yield in a Data Volcano Indexing Monitoring Titel van in P2P de Sensor presentatie DHT Networks 20

21 Fidelity and Yield in a Volcano Monitoring Sensor Network Geoff Werner-Allen, Konrad Lorincz, Jeff Johnson, Jonathan Lees and Matt Welsh OSDI 2006 October 19th, 2010 Duration: 20 min Kassandra Charalampidou Titel van de presentatie 21