Department or Program
Primary Wellesley Thesis Advisor
Benjamin P. Wood
Multithreading is a powerful model of parallel and concurrent programming. However, the presence of shared data leaves multithreaded programs vulnerable to concurrency errors such as data races, where two threads access and modify the same data concurrently and without synchronization. Data races lead to unpredictable program behavior and can be a source of data corruption. This work improves the precision of lockset-based dynamic data race detection without compromising soundness. Typically, lockset-based algorithms are sound but extremely imprecise. The algorithms presented in this work improve the precision of such algorithms by including thread-tracking information. Thread tracking helps detect patterns of intermittent thread-locality of shared data and eliminate false errors while still reporting all true errors. Experimental results show that thread-local analysis preserves soundness and improves precision of lockset-based data race detection by an average of 82%, with run-time slowdowns of less than 17%.