Candidate
Xi Cheng
Title
Anomaly Detection and Fault Localization Using Runtime State Models
Supervisor
Paul Ward
Abstract
Software failures are expensive, for some, even life endangering. Extensive research to provide software dependability through fault tolerance reduces the end-to-end visibility of failures, however, does not help in detecting faults or providing information to aid debugging. Efforts that focus on fault diagnosis either require manual input from the developers, work on a particular type of issues, i.e., performance issues, have coarse-grained granularities, i.e., on a per-component basis, or suffer from high false positives. Our work is the first to introduce the notion of Runtime State Models, a summarization of programs' behaviors through a collective view of variables and values, in the context of anomaly detection. Our evaluations show that Runtime State Models are effective in detecting software anomalies and are able to provide useful information in fault localization.