Interactive visualization of large scale scientific computations

高性能计算（HPC）的技术水平已经随着具有数千（如果不是数万的话）处理器的现代集群而快速发展。 大量的处理器和相应的大量主内存使研究人员能够在大量的时间步长上运行非常大的模型，可能会产生许多EB的数据。 这种庞大的数据量给传统的科学可视化方法带来了问题，在传统的科学可视化方法中，数据在计算过程中被收集在磁盘上，然后被传输到可视化工作站进行后期处理。 写入数据的过程现在消耗了大量的HPC资源，并且由于数据的大小，通过现有的计算机网络将其传输到可视化工作站不再可行。 Mercury系统通过在与科学计算相同的集群上执行至少部分可视化计算来支持非常大的模型的可视化。 这种可视化方法被称为原位可视化，直到最近才在研究界受到重视，但随着数据集大小的增加，它现在已成为一个重要的研究课题。 本项目研究软件体系结构和工具，在原位可视化和现有的可视化技术如何可以转移到这个新的环境。 水星系统是第一个通用的现场可视化系统之一，并将成为许多这些研究的基础。 该研究计划中产生的软件和技术将提高科学家和工程师使用的可视化系统的质量和效率。

The state of the art in high performance computing (HPC) has been advancing rapidly with modern clusters having thousands if not tens of thousands of processors. The large number of processors and the corresponding large amount of main memory allows researchers to run very large models over a large number of time steps potentially producing many exabytes of data. This large data volume causes problems for traditional approaches to scientific visualization where the data is collected on disk during the computation and then transferred to a visualization workstation for post processing. The process of writing the data is now consuming a significant amount of HPC resources and due to the size of the data transferring it over existing computer networks to a visualization workstation is no longer feasible. The Mercury system supports the visualization of very large models by performing at least part of the visualization computations on the same cluster as the scientific computation. This approach to visualization is called in situ visualization and until recently has not received much attention in the research community, but with increased data set size it has now become an important research topic. This project investigates software architectures and tools for in situ visualization and how existing visualization techniques can be transferred to this new environment. The Mercury system is one of the first general purpose in situ visualization systems and will form the basis for many of these studies. The software and techniques produced in this research program will improve both the quality and efficiency of the visualization systems used by scientists and engineers.