CDI-Type I: Small Resources Supercomputing: High Performance Computing in the Earth Sciences

CDI-Type I：小资源超级计算：地球科学中的高性能计算

• 批准号: 1027870
• 负责人: Greg Turk
• 金额: $ 51.2万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2010
• 资助国家: 美国
• 起止时间: 2010-10-01 至 2015-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1027870&HistoricalAwards=false
• 关键词: CDI; Type; Small; Resources; Supercomputing

There is a large body of scientific work on numerical simulation to model physical processes such as advection and diffusion in disparate fields such as ecology, oceanography, and climate science. Unfortunately, resolving the small time and spatial scale characteristics of these processes is computationally expensive and often limits fundamental studies to short times and small areas. Moreover, as these small-scale processes may lead to emergent properties on the regional and global scales, such simulations of limited scope (either in space or time) may lead to an incorrect understanding of large-scale dynamics. This is especially true in the climate sciences. In this work we propose to bring a user-friendly framework for High Performance Computing to the desktop computing environment so that small-scale processes can be simulated over large areas and the true nature of scaling understood. In particular, this work will take advantage of multi-processor Graphics Processing Units (GPUs) to increase the speed of computational simulations by up to three orders of magnitude over conventional workstation processors. These multi-processor GPUs provide a way to distribute the graphics rendering loads in such a way that, for example, state of the art computer games are visually realistic and fast enough to allow for user interactivity. Moreover, these same GPUs that permit fast graphics rendering can also be used to significantly accelerate numerical simulations. However, while there are individual groups that have taken advantage of GPU technology, by and large most scientists have not. Thus one goal of this work is to develop software tools for GPU simulation that are both flexible enough to use across a range of scientific problems, and are easy to use for researchers who have no GPU programming experience. This project will provide simulation tools for grid-based simulations for partial differential equations and particle-based simulations.While the technical objective of this work is to bring High Performance Computing to the desktop computing environment, it will be done so in the pursuit of scientific questions related to climate change. The investigators of this study have a history of expertise in the Arctic and the Everglades, and therefore will focus on problems relevant to these areas. The scientific goals are: (i) To better understand how processes scale. Specifically, whether small-scale non-linearities and feedbacks lead to emergent properties at scale. (ii) In the Arctic, the goal is to develop a mechanistic understanding of how snow and shrubs independently and interactively affect physical and biological controls over soil nitrogen dynamics and decomposition in arctic tundra, and how these dynamics in turn affect vegetation composition and productivity. (iii) In the Everglades, the goal is identify the dynamics that govern the formation and maintenance of the ridge and slough vegetation system and understand how management scenarios and changes in sea level and salt water intrusion will alter form and function of this patterned vegetation system.This research has the potential for broader impact in several areas. The investigations and discoveries about the interaction between vegetation, nutrients and water motion in the Everglades may provide guidance in terms of land use and resource management in this region. The snow capture models could provide a better understanding about the interaction between climate change and vegetation in the Arctic. Finally, the tools for grid-based simulations and particle-based simulations can provide a platform for using desktop computers with GPUs in a number of other areas of scientific investigation.

在生态学、海洋学和气候科学等不同领域，有大量关于数值模拟的科学工作来模拟物理过程，如平流和扩散。 不幸的是，解决这些过程的小时间和空间尺度的特点是计算昂贵的，往往限制了基础研究的时间短，面积小。 此外，由于这些小尺度过程可能导致区域和全球尺度上的突发性质，这种有限范围的模拟（无论是在空间还是时间上）可能导致对大尺度动态的错误理解。 在气候科学中尤其如此。 在这项工作中，我们提出了一个用户友好的框架，高性能计算的桌面计算环境，使小规模的过程可以模拟大面积和真正的性质扩展理解。 特别是，这项工作将利用多处理器图形处理单元（GPU）的优势，以提高计算模拟的速度高达三个数量级超过传统的工作站处理器。这些多处理器GPU提供了一种分配图形渲染负载的方式，例如，现有技术的计算机游戏在视觉上是逼真的，并且足够快以允许用户交互。 此外，这些允许快速图形渲染的GPU也可以用于显着加速数值模拟。 然而，虽然有个别团体已经利用了GPU技术，但大多数科学家都没有。 因此，这项工作的一个目标是开发用于GPU模拟的软件工具，这些工具既足够灵活，可以在一系列科学问题中使用，又易于没有GPU编程经验的研究人员使用。 该项目将为偏微分方程的网格模拟和粒子模拟提供模拟工具，虽然这项工作的技术目标是将高性能计算引入桌面计算环境，但这将是为了解决与气候变化有关的科学问题。 这项研究的调查人员在北极和大沼泽地的专业知识的历史，因此将重点放在与这些地区有关的问题。 科学目标是：（一）更好地了解过程如何扩展。 具体来说，小规模的非线性和反馈是否会导致大规模的涌现特性。(ii)在北极，我们的目标是开发一个机械的理解雪和灌木如何独立和交互影响物理和生物控制土壤氮动态和分解在北极苔原，以及这些动态反过来又如何影响植被组成和生产力。(iii)在大沼泽地，目标是确定的动态，管理的山脊和斯劳植被系统的形成和维护，并了解如何管理方案和海平面和盐水入侵的变化将改变这种模式的植被系统的形式和功能。对大沼泽地植被、营养盐和水分运动相互作用的研究和发现，可为该地区的土地利用和资源管理提供指导。雪捕获模型可以更好地了解北极气候变化与植被之间的相互作用。最后，基于网格的模拟和基于粒子的模拟工具可以为在许多其他科学研究领域使用具有GPU的台式计算机提供平台。