Detection and repair of software errors in hierarchical sensor networks

Sensor networks are being increasingly deployed for collecting critical data in various applications. Once deployed, a sensor network may experience faults at the individual node level or at an aggregate network level due to design errors in the protocol, implementation errors, or deployment conditions that are significantly different from the target environment. In many applications, the deployed system may fail to collect data in an accurate, complete, and timely manner due to such errors. If the network produces incorrect data, the resulting decisions on the data may be incorrect, and negatively impact the application. Hence, it is important to detect and diagnose these faults through run-time observation. Existing technologies face difficulty with wireless sensor networks due to the large scale of the networks, the resource constraints of bandwidth and energy on the sensing nodes, and the unreliability of the observation channels for recording the behavior. This paper presents a semi-automatic approach named H-SEND (hierarchical sensor network debugging) to observe the health of a sensor network and to remotely repair errors by reprogramming through the wireless network. In H-SEND, a programmer specifies correctness properties of the protocol ("invariants"). These invariants are associated with conditions (the "observed variables") of individual nodes or the network. The compiler automatically inserts checking code to ensure that the observed variables satisfy the invariants. The checking can be done locally or remotely, depending on the nature of the invariant. In the latter case, messages are generated automatically. If an error is detected at run-time, the logs of the observed variables are examined to analyze and correct the error. After errors are corrected, new programs or patches can be uploaded to the nodes through the wireless network. We construct a prototype to demonstrate the benefit of run-time detection and correction.