Visualizing Large Data Sets

可视化大型数据集

• 批准号: 9872147
• 负责人: William Schroeder
• 金额: $ 20.5万
• 依托单位: Rensselaer Polytechnic Institute
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1999
• 资助国家: 美国
• 起止时间: 1999-01-01 至 2001-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9872147&HistoricalAwards=false
• 关键词: Visualizing; Large; Data; Sets

Computer simulations are now an important means of understanding manycomplex physical phenomena. Numerical simulations are now vital instructural analysis, fluid dynamics, chemistry, and biomechanics for bothresearch and commercial purposes. Unfortunately, although parallelalgorithms and supercomputer systems can now handle the raw computations,visualization and data processing tools have failed to keep up with theseadvances. As a result, computers can generate a staggering amount of data- hundreds or thousands of gigabytes in time-dependent analysis - that muststill be translated to a form that humans can effectively understand.Without this translation, the value of simulation and analysis is sharplyreduced. This project will study new visualization methods andarchitectures needed to effectively apply numerical simulation tools.This project describes a data-flow architecture that can manage large datasets using computers ranging from small single-processor CPU's to high-endsupercomputers. The architecture minimizes disk access, maximizes use ofavailable memory, and lends itself to parallel algorithms. The approach isto stream coherent blocks of data between processes, and minimizing systemoverhead through careful use of configurable data caches and parsimoniousdata filters. This extends previous work by the investigators in an imageprocessing toolkit to handle more complex dataset types objects such asunstructured grids. The investigators will build their software as atoolkit of interoperable objects, and will make the software freelyavailable to other researchers.

计算机模拟现在是理解许多复杂物理现象的重要手段。 数值模拟现在对于研究和商业目的至关重要的结构分析、流体动力学、化学和生物力学。不幸的是，尽管并行算法和超级计算机系统现在可以处理原始计算，但可视化和数据处理工具却未能跟上这些进步。 因此，计算机可以生成数量惊人的数据（在与时间相关的分析中可达数百或数千 GB），这些数据仍必须转换为人类可以有效理解的形式。如果没有这种转换，模拟和分析的价值就会急剧降低。该项目将研究有效应用数值模拟工具所需的新可视化方法和架构。该项目描述了一种数据流架构，可以使用从小型单处理器 CPU 到高端超级计算机的计算机来管理大型数据集。该架构最大限度地减少了磁盘访问，最大限度地利用了可用内存，并适合并行算法。该方法是在进程之间传输连贯的数据块，并通过仔细使用可配置的数据缓存和简约的数据过滤器来最小化系统开销。这扩展了研究人员之前在图像处理工具包中的工作，以处理更复杂的数据集类型对象，例如非结构化网格。 研究人员将把他们的软件构建为可互操作对象的工具包，并将该软件免费提供给其他研究人员。