Dynamic datarace detection for object-oriented programs

Abstract

Multithreaded shared-memory programs are susceptible to dataraces, bugs that may exhibit themselves only in rare circumstances and can have detrimental effects on program behavior. Dataraces are often difficult to debug because they are difficult to reproduce and can affect program behavior in subtle ways, so tools which aid in detecting and preventing dataraces can be invaluable. Past dynamic datarace detection tools either incurred large overhead, ranging from 3x to 30x, or sacrificed precision in reducing overhead, reporting many false errors. This thesis presents a novel approach to efficient and precise datarace detection for multithreaded object-oriented programs. Our runtime datarace detector incurs an overhead ranging from 13% to 42% for our test suite, well below the overheads reported in previous work. Furthermore, our precise approach reveals dangerous dataraces in real programs with few spurious warnings.