CAREER: Autotuning Foundations for Exascale Computing

点击翻译按钮获取中文摘要

The goal of this research is to discover novel foundational principlesfor developing highly-efficient and reliable software that can achievesustainable performance on the exascale computing platforms expected by2020. Such platforms will deliver three orders of magnitude beyondtoday?s systems; harnessing this raw computational power couldrevolutionize our modeling and understanding of critical phenomena inareas like climate modeling, energy, medicine, sustainability,cosmology, engineering design, and massive-scale data analytics. Yet,developing software for exascale systems is a tremendous challengebecause the hardware is complex and it is not believed that the mostproductive ?high-level? software development environments (e.g.,programming languages and libraries) will be able to effectively exploitthese exascale systems.The investigator aims to address this challenge by using automatedtuning (autotuning) to eliminate the low performance traditionallyassociated with high-level programming models. This research (a)develops new model-driven frameworks for tuning parallel algorithms anddata structures, going beyond existing techniques that focus onlow-level code tuning; and (b) studies autotuning for programs expressedin high-level programming models, with the aim of eliminating theperformance gap. Concomitant with this research, the PI will create anew practicum course: The HPC Garage. The HPC Garage physicallyco-locates interdisciplinary teams in a social collaborative lab space;the teams engage in a year-long competition, called the XD Prize, todevelop highly scalable algorithms and software for NSF TeraGrid?snext-generation XD facilities. The HPC Garage also hosts summer internsin Georgia Tech?s Computing Research Undergraduate Intern SummerExperience (CRUISE) program, whose mission is to encourage students,especially those from underrepresented groups, to pursue graduatedegrees in computing.