FURI | Spring 2025

Detecting Data Races Using Enhanced Memory Protection Keys

Data races in multi-threaded programs can lead to unpredictable behavior, security vulnerabilities, and system failures. Existing dynamic race detection tools impose high overhead, making them impractical for large-scale applications. This research enhances the KARD race detection framework by integrating Extended Protection Keys (EPK), expanding protection domains from 16 to 7,680 to improve detection accuracy and scalability. Benchmark evaluations using PARSEC and SPLASH-2x will assess execution time and memory efficiency. The findings will contribute to cloud computing, aerospace, and financial systems, offering a scalable solution for secure, high-performance concurrent computing. Future research will focus on optimizing kernel-level synchronization mechanisms and OS-level concurrency models to further mitigate race conditions in modern multi-threaded architectures.