III: CGV: Small: A Scalable Visual Analytics Framework for Exascale Scientific Simulations

通过利用先进的并行计算系统，科学家可以回答对美国能源和经济安全至关重要的重要问题。艾级计算将进一步使科学家能够以更高的分辨率和更高的复杂性进行详细的模拟。为了研究各种物理、化学和生物现象，科学家需要以高交互性和保真度来探索海量和复杂的模拟数据。虽然可视化技术在最近几年有了很大的进步，但传统的可视化技术还没有为亿级系统和应用做好准备。未来的亿级系统预计将具有多核处理器、深度内存层次结构和高级别并发的特点。新的可视化技术的设计必须适应从复杂和极大的数据集及时发现的需要，以及这些新兴的硬件和软件趋势。该项目的目标是通过以整体方式用科学模拟调查可视化管道的完整过程来解决当前的技术差距，从而确保亿级数据可视化分析的并行性和效率。这个项目将把研究与教学和推广项目结合起来，在这些项目中，科学应用的可视化将被用作提高学生对科学和工程研究的兴趣和熟练程度的有效手段，并吸引和留住本科生和研究生，特别是女学生从事研究。该项目计划直接说明艾级计算视觉分析的关键组成部分之间的复杂相互依存关系。本项目的重点是三个方面的研究工作：(1)开发一种新的现场数据归约和索引算法，以获取大规模模拟中的关键信息；(2)研究并行可视化算法，以保证基于现场紧凑数据表示的大规模模拟数据的高通量和高分辨率探索的可扩展性能；(3)设计用户界面来解析和部署可视化分析的应用知识，以从现场模拟输出中获取关键的科学发现，并增强用户体验和性能。该项目是由真实世界的大型科学应用程序驱动的，这些应用程序涉及对具有不同数据类型的不断演变的现象进行建模和分析，并需要可扩展的可视分析功能。科学合作者将参与解决方案的开发、评估和部署，以缩小先进可视化技术和科学应用之间的差距，并帮助解决一些最具挑战性的科学问题。在该项目内开发的技术将很容易被许多应用程序采用，这些应用程序超出了具有类似需求的主要示范目标，因此将对科学家的数据分析和可视化能力产生重大影响。这项研究的成功可能会改变传统的科学发现渠道，并加快大规模模拟数据的研究。项目成果将通过在项目网站(http://cse.unl.edu/~yu/research/nsf15_exascale/).上公布的不同场所和形式进行传播

By leveraging advanced parallel computing systems, scientists can answer important questions that are critical to US energy and economic security. Exascale computing will further enable scientists to perform detailed simulations at higher resolution and greater complexity. Advanced visualization is necessary for scientists to explore massive and complex simulation data at high interactivity and fidelity to study various physical, chemical, and biological phenomena. Although visualization technology has significantly progressed in recent years, conventional visualization techniques are not yet ready for exascale systems and applications. Future exascale systems are expected to be characterized with many-core processors, deep memory hierarchies, and high levels of concurrency. The design of new visualization techniques must adapt to the need for timely discovery from complex and extremely large data sets as well as these emerging hardware and software trends. The goal of this project is to address the current technology gap by investigating a complete course of visualization pipeline with scientific simulations in a holistic fashion, and thus ensure parallelism and efficiency in exascale data visual analytics. This project will integrate research with teaching and outreach programs, where visualization of scientific applications will be used as an effective means to promote students' interest and proficiency in science and engineering studies, and to attract and retain both undergraduate and graduate students, particularly female students, into research.This project plans to account directly for the complex interdependencies with and among the critical components of visual analytics for exascale computing. This project focuses on three key research tasks: (1) developing a novel in-situ data reduction and indexing algorithm to capture essentials from large-scale simulations; (2) studying parallel visualization algorithms to promise scalable performance for high-throughput and high-resolution exploration of large-scale simulation data based on in-situ compact data representations; and (3) designing user interface to parse and deploy application knowledge for visual analytics to acquire critical scientific discovery from in-situ simulation output with enhanced user experience and performance. This project is driven by real-world large-scale scientific applications that involve the modeling and analysis of evolving phenomena with heterogeneous data types, and demand scalable capabilities of visual analytics. Scientific collaborators will be involved into the development, evaluation, and deployment of the solutions to close the gap between advanced visualization techniques and scientific applications, and help solve some of the most challenging scientific problems. The techniques developed within this project will be readily adapted for use by many applications beyond the primary demonstration targets with similar needs, and thus will have a significant impact on scientists' capability for data analysis and visualization. The success of this research will potentially change the conventional scientific discovery pipeline and accelerate the study of large-scale simulation data. The project results will be disseminated through different venues and forms that are publicized at the project website (http://cse.unl.edu/~yu/research/nsf15_exascale/).