Abstraction and Mining of Traces to Explain Concurrency Bugs

摘要: We propose an automated mining-based method for explaining concurrency bugs. use a data mining technique called sequential pattern to identify problematic sequences of concurrent read and write accesses the shared memory multi-threaded program. Our does not rely on any characteristics specific one type bug, thus providing general framework bug explanation. In our method, given set execution traces, we first mine that frequently occur in failing traces then rank them based number their occurrences passing traces. consider highly ranked events only explanation system failure, as they can reveal its causes Since scalability is limited by length present abstraction which shortens at cost introducing spurious explanations. Spurious well misleading explanations are eliminated subsequent filtering step, helping programmer focus likely failure. validate approach using case studies, including synthetic real-world