Discovering the Hidden Anomalies of Intermittent Computing

Energy harvesting battery-less embedded devices compute intermittently, as energy is available. Intermittent executions may differ from continuous ones due to repeated executions of non-idempotent code. This anomaly is normally recognized as a “bug” and solutions exist to retain equivalence between intermittent and continuous executions. We argue that our current understanding of these “bugs” is limited. We address this issue by devising techniques to comprehensively identify where and how intermittent and continuous executions possibly differ and by implementing them in SCEPTIC: a code analysis tool for intermittent programs. Thereby, we find execution anomalies and their manifested impact on program behavior in ways previously not considered. This analysis is enabled by SCEPTIC design, implementation, and performance. SCEPTIC runs up to ten orders of magnitude faster than the baselines we consider, enabling many types of analyses that would be otherwise impractical.