Detecting data races using dynamic escape analysis based on read barrier

Abstract

In multi-threaded programs, a data race results in extremely hard to locate bugs because of its nondeterministic behavior. This paper describes a novel dynamic data race detection method for object-oriented programming languages. The proposed method is based on the lockset algorithm. It uses read-barrier-based dynamic escape analysis for reducing number of memory locations that must be checked at runtime for detecting data races. We implemented the proposed data race detection method in HotSpot Java1 VM. The results of an experimental evaluation show a significant performance improvement over the previous write-barrier-based method and also that the proposed method can perform data race detection with a relatively small runtime overhead.