Output details
11 - Computer Science and Informatics
University of Edinburgh
Autotuning Skeleton-Driven Optimizations for Transactional Worklist Applications
<08>Originality: The first paper to apply a skeletal approach to software transactional memory (STM) systems.
Significance: STM systems offer a complex performing tuning space. Navigating this space manually negates much of the conceptual simplification offered by the TM programming model. We demonstrate that pattern specific constraints, carried implicitly by our worklist skeleton interface, allow the space to be traversed automatically in a machine and application sensitive way. This opens up a rich seam of research into the exploitation of other application patterns within TM.
Rigour: Real experiments with real benchmarks on real systems. Thorough investigation of underlying architectural performance factors.