SHF:Small: Collaborative Research: Tailoring Memory Systems for Data-Intensive HPC Applications

高性能计算对于美国的计算科学和工程具有重要的战略意义，并且在不断增加的失败中加速维持科学发现。为了将科学发现推进到一个新的水平，使新的科学任务能够及时完成，从整体上解决高性能计算平台中的内存性能问题至关重要。该项目为构建适合数据密集型应用（如大数据分析）的存储系统提供架构和系统支持和优化。该项目为本科生和研究生提供研究和教育机会，并在高性能计算领域培养新一代计算机科学家和工程师。本项目的目标是通过从几个方面设计新技术来解决构建高效存储系统的研究挑战。它在集群级开发了一种新的集中式内存刷新方案来管理内存刷新开销，这种开销对性能的影响和能耗越来越大。利用新的内存特性，设计了一种新的内存调度策略。它使处理器和操作系统可以使用内存特性/特性，以便它们可以做出明智的决定，以充分利用内存性能潜力。它利用内存计算来实现高效的原位处理。所有这些技术的集成为构建适合数据密集型应用的高效内存系统提供了一个整体解决方案。

High-performance computing is of strategic importance for computational science and engineering in the United States, and is on an accelerated path to sustaining scientific discovery at much increased flops. To advance the scientific discovery to the next level and allow new science missions to be accomplished in a timely manner, it is critical to address the memory performance holistically in high-performance computing platforms. This project provides architectural and system support and optimization for building memory systems tailored for data-intensive applications, e.g., big data analytics. This project offers research and educational opportunities for both undergraduate and graduate students, and trains a new generation of computer scientists and engineers in the area of high-performance computing. The objective of this project is to address the research challenges in building an efficient memory system by designing new techniques from several aspects. It develops a novel centralized memory refresh scheme at the cluster-level to manage memory refresh overhead, which has been increasingly performance-impacting and energy-consuming. It designs a new memory scheduling policy, taking advantages of new memory characteristics. It makes memory characteristics/peculiarities be available to the processor and operating system, so that they can make well-informed decisions to fully exploit memory performance potentials. It leverages in-memory computing to enable efficient in-situ processing. The integration of all these techniques provides a holistic solution to building an efficient memory system tailored for data-intensive applications.