II-New: System Acquisition for the Development of Scalable Parallel Algorithms for Scientific Computing

并行计算现在几乎在每一个科学和工程领域都扮演着核心角色。当今领先的超级计算机由数十万个处理器组成，预计在不久的将来将由数百万个处理器组成。特拉华大学的研究人员正在努力利用这种巨大的计算能力来取得突破性的科学发现。这项工作需要基础算法的研究和开发，以解决诸如多尺度耦合、可伸缩性和混合消息传递/线程编程等问题。研究人员正在为特拉华大学的科学计算研究和教学建立一个新的、高度并行的计算系统。这台计算机由大约1600个处理器组成，并且可以扩展到更多的处理器。对可扩展并行计算有共同兴趣的UD研究人员计划探索各种科学应用：用于环境湍流的3D Navier-Stokes求解器；海洋预报中的“集合分析”；计算电磁学;环境污染控制与材料合成中乳化液微观结构模拟原子和分子的电子结构计算；多尺度云动力学更可靠的天气和气候预报；计算碳纳米电子学；以及以涡度为主的气流。新计算机将用于优化每个应用程序中的当前算法，帮助确定可扩展性的限制，并探索新的想法，以实现扩展到数十万个处理器并利用混合架构的优势。该系统促进了对应用问题感兴趣的教师和对计算机科学有专业知识的教师之间的合作。有效的并行数据再分配和并行代码的自动优化库是参与该项目的计算机科学家增强科学应用的一些方法。

Parallel computing now plays a central role in nearly every science and engineering endeavor. The leading supercomputers today comprise hundreds of thousands of processors, and in the near future are expected to comprise millions of processors. Researchers at the University of Delaware are working to harness this vast computing power to make ground-breaking scientific discoveries. This work requires fundamental algorithmic research and development to address issues such as multiscale coupling, scalability, and hybrid message-passing/threaded programming.The investigators are building a new, highly parallel, computing system for scientific computation research and teaching at the University of Delaware. This computer consists of approximately 1,600 processors, and is expandable to many more. UD researchers who share a common interest in scalable parallel computation plan to explore a variety of scientific applications: 3D Navier-Stokes solvers for environmental turbulent flows; "ensemble analysis" in oceanic forecasting; computational electromagnetism; simulation of emulsion microstructure for environmental pollution control and material synthesis; electronic structure calculations for atoms and molecules; multiscale cloud dynamics for more reliable weather and climate prediction; computational carbon nanoelectronics; and vorticity-dominated flows. The new computer will be used to optimize the current algorithms in each application, help identify limitations to scalability, and explore new ideas to enable scaling to hundreds of thousands of processors and to take advantage of hybrid architectures. The system motivates collaboration among faculty interested in application problems and faculty with expertise in computer science. Libraries for efficient parallel data redistribution and automatic optimization of parallel codes are some of the ways the computer scientists participating in the project are enhancing the scientific applications.