Mining tree-structured data on multicore systems

摘要: Mining frequent subtrees in a database of rooted and labeled trees is an important problem many domains, ranging from phylogenetic analysis to biochemistry linguistic parsing XML data analysis. In this work we revisit develop architecture conscious solution targeting emerging multicore systems. Specifically identify sequence memory related optimizations that significantly improve the spatial temporal locality state-of-the-art sequential algorithm -- alleviating effects latency. Additionally, these are shown reduce pressure on front-side bus, consideration context large-scale architectures. We then demonstrate while necessary not sufficient for efficient parallelization multicores, primarily due parametric data-driven factors which make load balancing significant challenge. To address challenge, present methodology adaptively automatically modulates type granularity being shared among different cores. The resulting achieves near perfect parallel efficiency up 16 processors challenging real world applications. have general purpose utility key out-come development scheduling service moldable task