DEVELOPMENT OF DYNAMIC DATA RACE DETECTION FOR CONCURRENT SOFTWARE SYSTEMS

Concurrent programs are prone to data races, which are notoriously difficult to detect and reproduce. Static analysis often yields excessive false positives, while traditional dynamic testing can miss races due to specific thread interleavings. Existing dynamic detectors struggle with the overhead of precise analysis and the scalability for large, long-running applications. This creates a critical need for efficient, on-the-fly race detection techniques that are both precise and practical. Our methodology employs a hybrid approach of happens-before and lockset analysis for precise on-the-fly detection. We instrument program binaries to monitor memory accesses and synchronization operations during execution. A vector clock algorithm efficiently tracks potential causal relationships between threads. This allows for the immediate flagging of conflicting, unordered accesses as data races. Our dynamic detector successfully identified 15 true data races in the benchmark suite, with zero false positives. It incurred an average runtime overhead of 25% and a memory overhead of 45%, demonstrating practical performance. The tool also successfully localized the precise instruction pairs and variable involved in each race. These results validate our hybrid analysis as both precise and efficient for real-world applications. Our study confirms that hybrid happens-before and lockset analysis is highly effective for precise and practical dynamic race detection. We recommend its integration into continuous testing pipelines to catch concurrency bugs early in the development cycle. Future work should focus on reducing memory overhead and extending detection to more relaxed memory models. Widespread adoption of such tools is crucial for improving the reliability of concurrent software