CSR: Large: VarSys: Managing Variability in High-Performance Computing Systems

Exascale高性能计算（HPC）系统需要满足21世纪科学计算和电子商务的许多重大挑战。这些系统的可变性越来越棘手，并导致性能损失和能源浪费。这个项目正在构建可变性系统（VarSys）软件框架，以基础工程原理为基础，在大型高级计算系统中实现可变性管理。这项工作中的智力挑战包括确定可变性可以管理的程度，以及由这种管理产生的权衡。该工作将演示在大规模系统和云基础设施上使用可变性管理。该项目的继续将努力创造方法，以建立一个社区生态系统，以基准进展，并使更广泛的社区参与这项研究。目前，有超过25名教职员工和学生（包括6名女性和2名少数民族学生）参与了该项目，200多名学生在操作系统课程中接触到系统变异性。该项目还将制作额外的引人注目的视频和其他媒体文物，以描述进展并捕捉研究中固有的兴奋。关键词：高性能计算；统计和数学建模；该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Exascale high-performance computing (HPC) systems are required to meet the demands of many grand challenges for scientific computing and e-commerce in the 21st century. Variability in these systems is growing intractably and leads to performance loss and energy waste. This project is building the Variability System (VarSys) software framework, grounded in fundamental engineering principles, to enable variability management in large-scale advanced computing systems. The intellectual challenges in this work include determining the extent to which variability can be managed and the tradeoffs that result from such management. The work will demonstrate the use of variability management on large scale systems and cloud infrastructures. The continuation of the project will endeavor to create methodologies to establish a community ecosystem to benchmark progress and engage a broader community in this research. Currently, there are over 25 faculty, staff and students (including 6 women and 2 minority students) participating in the project and more than 200 students have been exposed to system variability in the operating system courses. The project will also create additional compelling videos and other media artifacts to describe the progress and capture the excitement inherent to the research.Keywords: high-performance computing; statistical and mathematical modeling; variability-performance trade-offsThis award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.