Abstracts Category : Other

Error propagation analysis of multithreaded programs using likely invariants

by Abraham Chan

Error Propagation Analysis (EPA) is a technique for understanding how errors affecta programs execution and result in program failures. For this purpose, EPAusually compares the traces of a fault-free (golden) run with those from a faultyrun of the program. This makes existing EPA approaches brittle for multithreadedprograms, which do not typically have a deterministic golden run. In this thesis, westudy the use of likely invariants generated by automated approaches as alternativesfor golden run based EPA in multithreaded programs. We present Invariant PropagationAnalysis (IPA), an approach and a framework for automatically derivinginvariants for multithreaded programs, and using the invariants for EPA.We evaluatethe invariants derived by IPA in terms of their coverage for different fault typesacross six representative programs through fault injection experiments. We findthat stable invariants can be inferred in all six programs, although their coverage offaults depends on the application and the fault type.