EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

ThreadMonitor : Low‐Overhead Data Race Detection Using Intel Processor Trace

Farzam Dorostkar, Michel Dagenais, Ankush Tyagi, Vince Bridgers

Concurrency and Computation: Practice and Experience · 2026-01 · 38 (2)

DOI: 10.1002/cpe.70517 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Concurrency and Computation: Practice and Experience, online erschienen am 21.01.2026. Verfasst von Farzam Dorostkar und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Data races are among the most difficult multithreading bugs to find, due to their non‐deterministic nature. This and the increasing popularity of multithreaded programming have led to the need for practical automated data race detection. In this context, dynamic data race detectors have received more attention, compared to static tools, owing to their higher accuracy and scalability. Yet, state‐of‐the‐art dynamic data race detectors cannot be used in many real‐world testing scenarios, since they cause significant slowdown and memory overhead. Notably, ThreadSanitizer (TSan), the default dynamic data race detector in both clang and gcc compilers, is reported to typically impose a – slowdown and a – memory overhead, which is not tolerable in many industrial use cases. To address this issue, this paper introduces ThreadMonitor (TMon), a low‐overhead postmortem data race detector for multithreaded C/C++ programs that use the Pthread library. At runtime, TMon traces the information required for detecting occurrences of data races (i.e., shared memory accesses and timing constraints among threads) using Intel Processor Trace (Intel PT), a non‐intrusive hardware feature dedicated to tracing software execution. Thereafter, its postmortem analyzer examines the collected trace data to determine whether the traced program execution exhibited data races, performing a verification similar to that carried out by TSan at runtime. Introducing algorithmic improvements in its postmortem analyzer, TMon can further achieve a higher data race detection coverage compared to TSan. TMon has no direct data memory overhead, incurs minimal instruction memory overhead, and causes a very small slowdown, making it an ideal choice in test environments with limited resources.

Zitieren

APA

Dorostkar, F., Dagenais, M., Tyagi, A., & Bridgers, V. (2026). ThreadMonitor : Low‐Overhead Data Race Detection Using Intel Processor Trace. Concurrency and Computation: Practice and Experience, 38(2). https://doi.org/10.1002/cpe.70517

Vancouver

Dorostkar F, Dagenais M, Tyagi A, Bridgers V. ThreadMonitor : Low‐Overhead Data Race Detection Using Intel Processor Trace. Concurrency and Computation: Practice and Experience. 2026;38(2). doi:10.1002/cpe.70517

BibTeX
@article{Dorostkar2026,
  author = {Dorostkar, Farzam and Dagenais, Michel and Tyagi, Ankush and Bridgers, Vince},
  title = {ThreadMonitor : Low‐Overhead Data Race Detection Using Intel Processor Trace},
  journal = {Concurrency and Computation: Practice and Experience},
  year = {2026},
  volume = {38},
  number = {2},
  doi = {10.1002/cpe.70517}
}
Kostenlos starten →