Grammar-aware Parallelization for Scalable XPath Querying


Semi-structured data emerge in many domains, especially in web analytics and business intelligence. However, querying such data is inherently sequential due to the nested structure of input data. Existing solutions pessimistically enumerate all execution paths to circumvent dependencies, yielding sub-optimal performance and limited scalability. This paper presents GAP, a parallelization scheme that, for the first time, leverages the grammar of the input data to boost the parallelization efficiency. GAP leverages static analysis to infer feasible execution paths for specific con- texts based on the grammar of the semi-structured data. It can eliminate unnecessary paths without compromising the correctness. In the absence of a pre-defined grammar, GAP switches into a speculative execution mode and takes potentially incomplete grammar extracted either from prior inputs. Together, the dual-mode GAP reduces the execution paths from all paths to a minimum, therefore maximizing the parallelization efficiency and scalability. The benefits of path elimination go beyond reducing extra computation -- it also enables the use of more efficient data structures, which further improves the efficiency. An evaluation on a large set of standard benchmarks with diverse queries shows that GAP yields significant efficiency increase and boosts the speedup of the state-of-the-art from 2.9X to 17.6X on a 20-core ma- chine for a set of 200 queries.

DOI: 10.1145/3018743.3018772

15 Figures and Tables

Cite this paper

@inproceedings{Jiang2017GrammarawarePF, title={Grammar-aware Parallelization for Scalable XPath Querying}, author={Lin Jiang and Zhijia Zhao}, booktitle={PPOPP}, year={2017} }