TY - GEN
T1 - SNTS
T2 - 3rd IEEE International Conference on Distributed Computing in Sensor Systems, DCOSS 2007
AU - Khan, Mohammad Maifi Hasan
AU - Luo, Liqian
AU - Huang, Chengdu
AU - Abdelzaher, Tarek
PY - 2007
Y1 - 2007
N2 - Sensor network troubleshooting is a notoriously difficult task, further exacerbated by resource constraints, unreliable components, unpredictable natural phenomena, and experimental programming paradigms. This paper presents SNTS (Sensor Network Troubleshooting Suite), a tool that performs automated failure diagnosis in sensor networks. SNTS can be used to monitor network conditions using simple visualization techniques as well as to troubleshoot deployed distributed sensor systems using data mining approaches. It is composed of (i) a data collection front-end that records events internal to the network and (ii) a data processing back-end for subsequent analysis. We use data mining techniques to automate failure diagnosis on the back-end. The assumption is that the occurrence of execution conditions that cause failures (e.g., traversal of an execution path that contains a "bug" or occurrence of a sequence of events that a protocol was not designed to handle) will have a measurable correlation (by causality) with the resulting failure itself. Hence, by mining for network conditions that correlate with failure states the root causes of failure are revealed with high probability. To evaluate the effectiveness of the tool, we have used it to troubleshoot a tracking system called EnviroTrack [4], which, although performs well most of the time, occasionally fails to track targets correctly. Results show that SNTS can identify the major causes of the problem and give developers useful hints on improving the performance of the tracking system.
AB - Sensor network troubleshooting is a notoriously difficult task, further exacerbated by resource constraints, unreliable components, unpredictable natural phenomena, and experimental programming paradigms. This paper presents SNTS (Sensor Network Troubleshooting Suite), a tool that performs automated failure diagnosis in sensor networks. SNTS can be used to monitor network conditions using simple visualization techniques as well as to troubleshoot deployed distributed sensor systems using data mining approaches. It is composed of (i) a data collection front-end that records events internal to the network and (ii) a data processing back-end for subsequent analysis. We use data mining techniques to automate failure diagnosis on the back-end. The assumption is that the occurrence of execution conditions that cause failures (e.g., traversal of an execution path that contains a "bug" or occurrence of a sequence of events that a protocol was not designed to handle) will have a measurable correlation (by causality) with the resulting failure itself. Hence, by mining for network conditions that correlate with failure states the root causes of failure are revealed with high probability. To evaluate the effectiveness of the tool, we have used it to troubleshoot a tracking system called EnviroTrack [4], which, although performs well most of the time, occasionally fails to track targets correctly. Results show that SNTS can identify the major causes of the problem and give developers useful hints on improving the performance of the tracking system.
KW - Data mining
KW - Distributed troubleshooting
KW - Sensor network
UR - http://www.scopus.com/inward/record.url?scp=38149014137&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=38149014137&partnerID=8YFLogxK
U2 - 10.1007/978-3-540-73090-3_10
DO - 10.1007/978-3-540-73090-3_10
M3 - Conference contribution
AN - SCOPUS:38149014137
SN - 3540730893
SN - 9783540730897
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 142
EP - 157
BT - Distributed Computing in Sensor Systems - Third IEEE International Conference, DCOSS 2007, Proceedings
PB - Springer
Y2 - 18 June 2007 through 20 June 2007
ER -