Goldilocks: Efficiently Computing the Happens-Before Relation Using Locksets

摘要: We present a new lockset-based algorithm, Goldilocks, for precisely computing the happens-before relation and thereby detecting data-races at runtime. Dynamic race detection algorithms in literature are based on vector clocks or locksets. Vector-clock-based compute but have significantly more overhead. Previous algorithms, other hand, imprecise. They check adherence to particular synchronization discipline, i.e., sufficient condition freedom may generate false warnings. Our like clocks, is precise, yet it efficient since purely lockset based. We implemented our algorithm inside Kaffe Java Virtual Machine. implementation incorporates lazy evaluation of locksets certain “short-circuit checks” which contribute its efficiency. Experimental results indicate that algorithm's overhead much less than vector-clock very close Eraser algorithm.