Non-intrusive on-the-fly data race detection using execution replay

TL;DR

This paper introduces a practical, non-intrusive method for detecting data races in parallel programs by combining execution replay with on-the-fly detection, minimizing overhead and avoiding Heisenbugs.

Contribution

It presents a novel integration of execution replay with real-time data race detection that operates with minimal performance impact and memory usage.

Findings

Enables continuous, non-intrusive data race detection during execution.

Uses multilevel bitmaps and snooped matrix clocks to reduce memory overhead.

Eliminates Heisenbugs by allowing tracing to run constantly.

Abstract

This paper presents a practical solution for detecting data races in parallel programs. The solution consists of a combination of execution replay (RecPlay) with automatic on-the-fly data race detection. This combination enables us to perform the data race detection on an unaltered execution (almost no probe effect). Furthermore, the usage of multilevel bitmaps and snooped matrix clocks limits the amount of memory used. As the record phase of RecPlay is highly efficient, there is no need to switch it off, hereby eliminating the possibility of Heisenbugs because tracing can be left on all the time.