Collaborative Research: High-Performance Particle Simulations from Desktops to Exaflops

现代高性能计算机体系结构强调并行性、异构性和可重新配置性，以提供性能和功率效率的增强。这些复杂程度给计算物理学家带来了新的挑战，他们需要重新映射现有的软件，以便模拟工具在新的计算机体系结构上有效地运行。特别是，使用粒子池（PIC）方法的等离子体模拟被广泛应用于各种主题，包括惯性聚变能、基于等离子体的加速、空间天气和天体物理学。该项目的目标是开发一组通用骨架代码，可用于实验、教育和为下一代多核体系结构开发适应性强的PIC软件。这里支持的活动将集中于将最佳软件实践纳入生产代码，供当前和未来的高性能计算机使用，试验负载平衡和性能和准确性折衷的新想法，分析和存储大型数据集，以及开发PIC代码的“桌面”版本，供广泛受众使用并扩大其影响。这个项目将为科学家提供强大的模拟工具，利用视频游戏中常用的图形卡的巨大计算能力。这些工具将使科学家能够从事等离子体物理学的研究，建造紧凑的粒子加速器，了解宇宙射线，并使核聚变成为人类实用的清洁能源。这里开发的工具也将作为学生学习等离子体物理复杂方面的教学工具。该项目将为追求物理学事业的学生提供独特的培训和研究机会，并为他们提供适合现代高性能计算平台的最先进技术的实践经验。

Modern high-performance computer architectures highlight parallelism, heterogeneity, and re-configurability to provide both performance and power efficiency enhancements. These degrees of complexity present new challenges to computational physicists, who are required to remap existing software so that simulation tools run efficiently on novel computer architectures. In particular, plasma simulations using the particle-in-cell (PIC) method are widely used on a variety of topics including inertial fusion energy, plasma-based acceleration, space weather, and astrophysics. The goal of this project is to develop a set of general skeleton codes that can be used for experimentation, education, and for developing adaptable PIC software for next-generation many-core architectures. The activities supported here will focus on incorporating the best software practices into production codes for use on present and future high-performance computers, on experimenting with novel ideas for load balancing and tradeoffs in performance and accuracy, for analyzing and storing large data sets, and on developing "desktop" versions of PIC codes for use by a wide audience and broaden their impact. This project will provide scientists with powerful simulation tools by tapping into the vast computing power of graphic cards commonly used for video games. These tools will enable scientists to pursue research in plasma physics, to build compact particle accelerators, to understand cosmic rays, and to make fusion a practical clean energy source for mankind. The tools developed here will also serve as teaching tools for students to learn complex aspects of plasma physics. This project will provide unique training and research opportunities for students pursuing a career in physics and hands-on experience with state-of-the-art techniques suitable for modern high-performance computational platforms.