Detecting atomic-set serializability violations in multithreaded programs through active randomized testing

• Published in: 2010 ACM/IEEE 32nd International Conference on Software Engineering Vol. 1; pp. 235 - 244
• Main Authors: Lai, Zhifeng, Cheung, S. C., Chan, W. K.
• Format: Conference Proceeding
• Language: English; Japanese
• Published: New York, NY, USA ACM 01.05.2010; IEEE
• Series: ACM Conferences
• Subjects: atomicity; Computer bugs; Concrete; Concurrent computing; dynamic analysis; General and reference > Cross-computing tools and techniques > Reliability; Instruction sets; Monitoring; serializability; Software and its engineering > Software creation and management > Software verification and validation > Software defect analysis > Software testing and debugging; Software and its engineering > Software organization and properties > Extra-functional properties > Software reliability; software testing; Synchronization; Testing; serializability; dynamic analysis; atomicity; software testing
• ISBN: 9781605587196; 1605587192
• ISSN: 0270-5257
• DOI: 10.1145/1806799.1806836

Concurrency bugs are notoriously difficult to detect because there can be vast combinations of interleavings among concurrent threads, yet only a small fraction can reveal them. Atomic-set serializability characterizes a wide range of concurrency bugs, including data races and atomicity violations. In this paper, we propose a two-phase testing technique that can effectively detect atomic-set serializability violations. In Phase I, our technique infers potential violations that do not appear in a concrete execution and prunes those interleavings that are violation-free. In Phase II, our technique actively controls a thread scheduler to enumerate these potential scenarios identified in Phase I to look for real violations. We have implemented our technique as a prototype system AssetFuzzer and applied it to a number of subject programs for evaluating concurrency defect analysis techniques. The experimental results show that AssetFuzzer can identify more concurrency bugs than two recent testing tools RaceFuzzer and AtomFuzzer.