Scalable and precise dynamic datarace detection for structured parallelism

Abstract

Existing dynamic race detectors suffer from at least one of the following three limitations: (i) space overhead per memory location grows linearly with the number of parallel threads [13], severely limiting the parallelism that the algorithm can handle; (ii) sequentialization : the parallel program must be processed in a sequential order, usually depth-first [12, 24]. This prevents the analysis from scaling with available hardware parallelism, inherently limiting its performance; (iii) inefficiency : even though race detectors with good theoretical complexity exist, they do not admit efficient implem entations and are unsuitable for practical use [4, 18]. We present a new precise dynamic race detector that leverages structured parallelism in order to address these limitations. Our algorithm requires constant space per memory location, works in parallel, and is efficient in practice. We implemented and evaluated our algorithm on a set of 15 benchmarks. Our experimental results indicate an average (geometric mean) slowdown of 2.78x on a 16-core SMP system.

Dynamic race prediction in linear time

“…Tools such as ROADRUN-NER [15] and Sofya [21] provide frameworks for implementing dynamic analysis tools. Techniques such as [7,12,32,43,48] leverage structured parallelism to optimize memory overhead for dynamic race detection.…”

Dynamic enforcement of determinism in a parallel scripting language

“…Mellor-Crummey's offset-span labeling [23] provides a space-efficient alternative to vector clocks for fine-grain split-merge programs, but was also implemented sequentially. The state of the art would appear to be the Habañero Java SPD3 detector [27], which runs in parallel and provides precise (sound and complete) data race detection for arbitrary splitmerge (async-finish) programs. 2 Like most recent race detectors, SPD3 relies on shadow memory to store metadata for each shared memory location.…”

“…For these we compare TARDIS, in both default and detail mode, to besteffort reimplementations of the Cilk Nondeterminator [11] and SPD3 [27] race detectors. Both prior systems are shadow-memory based, and do not record source-code-level information.…”

Access Annotation for Safe Program Parallelization

“…On-the-fly race detection can be done efficiently for perfectly nested fork-join parallelism [17,20]. Callahan and others showed that at the program level, the problem of post-wait race checking is co-NP hard and gave an approximate solution based on dataflow analysis [7].…”

“…As discussed in Section 4, BOP needs at most two annotations per datum per task and does not synchronize at every access. Race checkers, like DSTM mentioned in Section 4, use shared memory and have to monitor all accesses, and except for the recent Tardis system [13], update a shared data structure at each shared-data access [20].…”