| Abstract: |
Health monitoring is an integral service for the long-term operation of wireless sensor networks. Instead of actively adding probe or status traffic, recently proposed passive health
monitoring systems infer system health solely from existing application traffic. Our results from extensive testbed experiments prove that the detection rate of an exemplary monitoring
application is significantly improved, i.e., by up to 28% to overall 94%, when a passive health estimation method is complemented with a minimally active component. This paper presents Hybrid
Monitoring, a novel health monitoring system that combines the advantages of passive and active health monitoring systems. Results from the analysis of passively reconstructed, inexact per-hop
timing information are significantly improved when one bit of extra information is added to every packet. The resulting system is able to estimate a significant health metric, i.e., the number of
occurred failure events on a node, with a high confidence. The temporal resolution of the estimated signal is equal to the packet sampling interval, the performance of our system is not affected
by large packet delays, e.g., when a sensor node is disconnected for a long time. |
