PECASE: Parallel Visualization and Interaction Techniques for Exploring Large Scale Volume Data

向用户展示大型科学数据集的最佳方法之一是可视化。在过去的十年里，由于3D图形硬件和软件技术的进步，我们创建有用的可视化效果的能力经历了一场革命。然而，随着数据的复杂性和规模不断增长，无论是来自物理实验还是计算机模拟，我们仍然面临着无穷无尽的挑战。在许多情况下，大规模数据集的可视化是非常昂贵的，有时甚至比最初产生数据的超级计算机模拟还要昂贵。本项目将针对这一情况产生的两个问题进行全面研究。首先，我们可以使用并行超级计算机来生成数据并以最佳方式进行可视化计算吗？第二，是否有机制和方法来提高科学家进行勘探和生产可视化的生产力？该项目将开发一个完全并行的可视化系统，以最大限度地提高系统的总体吞吐量。该系统将强调在PC集群上进行计算。该项目这一部分的四个主要子主题将是用于准备可视化数据的并行预处理算法、用于生成图形的并行渲染算法、用于高效管理图形的并行图像压缩和传输以及用于存储用于可视化的数据的应用程序控制的并行I/O。该项目将特别侧重于并行特征增强的可视化算法和可视化数据的并行子集。与该项目的研究部分相辅相成的是针对本科生和研究生的教育部分，旨在通过实践经验激发兴趣和培养创造力。这种教育的一个方面是为非CS研究生开设一门应用可视化课程，这应该会促进学生家庭部门之间更多的互动和思想的交流。

One of the best methods of presenting large scientific data sets to users is visualization. Over the past ten years, our ability to create useful visualizations has undergone a revolution due to advances in 3D graphics hardware and software technology. However, we still face endless challenges as the complexity and scale of the data, whether from physical experiments or computer simulation, continue to grow. In many cases, visualization of large-scale data sets is very expensive, sometimes even more expensive than the supercomputer simulation that produced the data in the first place. This project will perform a comprehensive study aimed at two questions that arise from this situation. First, can we use parallel supercomputers to generate data and for visualization calculations in an optimal fashion? Second, are there mechanisms and methodologies to enhance the productivity of scientists doing exploratory and production visualization? The project will develop a fully parallel visualization system to maximize the overall throughput of the system. This system will emphasize computing on clusters of PCs. Four major sub-themes of this part of the project will be parallel preprocessing algorithms to prepare data for visualization, parallel rendering algorithms to produce the graphics, parallel image compression and transport to efficiently manage the graphics, and application-controlled parallel I/O to store the data for visualization. The project will particularly focus on parallel feature-enhanced visualization algorithms and parallel subsetting of the visualized data. Complementing the research component of this project is an educational component, directed at undergraduate and graduate students, aimed at inspiring interest and fostering creativity through hands-on experiences. One aspect of this education is an applied visualization course to non-CS graduate students, which should foster greater interaction and cross-fertilization of ideas from the students' home departments.