Noise injection heuristics for concurrency testing

Testing of concurrent software is difficult due to the non-determinism present in scheduling of concurrent threads. Existing testing approaches tackle this problem either using a modified scheduler which allows to systematically explore possible scheduling alternatives or using random or heuristic noise injection which allows to observe different scheduling scenarios. In this paper, we experimentally compare several existing noise injection heuristics both from the point of view of coverage of possible behaviours as well as from the point of view of error discovery probability. Moreover, we also propose a new noise injection heuristics which uses concurrency coverage information to decide where to put noise and show that it can outperform the existing approaches in certain cases.