Динамические средства обнаружения гонок в параллельных программах

Abstract

One of the most hazardous and hardly reproducible errors that occur in multithreaded programs are data races — unsynchronized accesses to same shared memory fragment from several threads, where one access is write. Generally data races are weakly localized and damage global data structures. Manual detection of data races is very complicated. There was a lot of research in this area, but automatic data race detection remains an actual issue. In this review evolution of existing approaches is considered and their advantages and drawbacks are analyzed. Special attention is paid to automatic race detection in Java applications.