CRII: CSR: MPI-ACC_GIS: Accelerating Geo-Spatial Computations on HPC Platform

空间计算是许多科学和工程应用中使用的拓扑和几何分析的核心。空间数据用于地图导航应用程序。公共卫生、城市规划、交通、科学界和私营部门的政府机构依靠空间数据挖掘和分析来获得见解并制定可行的计划。然而，随着空间数据的规模和复杂性的增加，传统的基于桌面的分析已经不能满足任务的需要，需要在现代计算机和超级计算机上加速计算密集型的空间应用，以实时获得结果。这需要新的软件设计和实现，以及有效的空间查询、连接和覆盖算法，这些算法可以随着数据和可用硬件资源的扩展而扩展。本计画的第一个目标是为计算几何问题开发实用的基于平面扫描的并行算法。第二个目标是将空间数据感知注入到现有的基于消息传递接口（MPI）的地理信息系统（GIS）中，并通过使用专门的软件包整合新功能来利用高性能计算资源。将开发一种新的负载平衡技术，该技术通过MPI库支持的非连续文件读取在文件分区阶段工作。MPI-Vector-IO库将在存储在并行文件系统上的大型多边形数据集上实现高效的并行输入/输出，从而提高技术水平。MPI-ACC-GIS软件将使现有的顺序工具能够在具有数千个核心的高性能计算环境中使用。基于并发数据结构的平面扫描算法在共享内存机器上的并行化将影响许多依赖平面扫描的计算几何算法的高效实现。MPI-ACC-GIS软件的开发将对地理空间数据科学家有用。此外，该项目在培养本科生和研究生进行高性能计算研究方面具有更广泛的影响。MPI-ACC-GIS软件实施新的算法和方法，以及测试数据、结果和基准将在大学主办的网站上公开提供。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Spatial computations are at the heart of topological and geometric analysis used in many science and engineering applications. Spatial data is used in mapping applications for navigation. Government agencies in public health, urban planning, transportation, scientific communities and private sector depend on spatial data mining and analysis to gain insights and produce actionable plans. However, with the increase in the size and complexity of spatial data, traditional desktop-based analysis is inadequate for the task, and there is a need to accelerate compute-intensive spatial applications on modern computers and supercomputers to get results in real-time. This requires new software design and implementation, as well as efficient algorithms for spatial query, join and overlay that can scale with the data and the available hardware resources. The first objective of this project is to develop practical parallel algorithms based on plane sweep for computational geometry problems. The second objective is to inject spatial data awareness into the existing Message Passing Interface (MPI)-based Geographic Information System (GIS), and incorporate new features to leverage high performace computing resources by using specialized software packages. A new load balancing technique will be developed that works during the file partitioning phase by non-contiguous file reading supported by the MPI library.The MPI-Vector-IO library will enable efficient parallel input/output on large polygonal datasets stored on parallel filesystems, thus improving the state-of-the-art. The MPI-ACC-GIS software will enable existing sequential tools to be used in a high performance computing environment with thousands of cores. Parallelization of plane sweep algorithm based on concurrent data structures on shared memory machines will impact many computational geometry algorithms that rely on plane sweep for efficient implementation. Development of MPI-ACC-GIS software will be useful to geo-spatial data scientists. In addition, the project has a broader impact in training undergraduate and graduate students to perform research in high performance computing.MPI-ACC-GIS software implementing new algorithms and methodologies along with test data, results and benchmarks will be made publicly available on a website hosted at the university.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.