SDCI HPC: Improvement and Release of the Uintah Computational Framework

智能优点：Project建议改进和扩展Uintah框架，以支持在具有数万个节点和复杂的通信/内存层次结构的现代HPC架构上进行自适应计算。犹他州是在犹他大学意外火灾和爆炸模拟中心(C-SAFE)下开发的，C-SAFE是美国能源部资助的学术联盟项目。这项提议旨在将犹他州提供的独特能力带给更广泛的NSF科学和工程界。同时，所采用的基本方法将说明如何以容错和自适应的方式在大规模高性能计算体系结构上使用这些软件包。Uintah适用于需要流体-结构相互作用和高度变形模型的广泛工程领域，并用于目前高级测试版形式，用于犹他大学由NSF资助的生物工程和计算机科学项目(如组织建模)，以及犹他州和其他机构的其他NIH、DARPA、DOE和行业资助的项目(如心脏损伤建模和冲击波模拟)。这个提议的项目的目的是将这个软件转移到更广泛的NSF计算科学界。Uintah是通过麻省理工学院开源许可证获得的开源软件。Uintah将扩展为通过冗余计算提供额外的容错能力，这将得到Uintah系统基于组件的架构的便利。我们将向Uintah添加分层蒙特卡罗辐射传输模型，以提供额外的热传输和电磁波传播能力。最后，我们将演示Uintah on Petascale架构在现实世界工程设计问题上的可扩展性，并通过在线维基等现代协作工具支持Uintah的外部用户进行NSF资助的研究和其他研究。这个项目有两个主要的预期结果。第一个直接结果是Uintah的增强版本，有可能扩展到数千个处理器，并开始接近Petaflop性能。该代码将立即用于许多涉及危险的应用程序的计算建模。第二个贡献不那么直接，但同样重要。Uintah内部底层任务图方法的使用将提供一个实例，说明如何以容错和自适应的方式将潜在的重要方法应用于大规模并行。作为上述工作的一部分，其他贡献将包括在流体-结构相互作用问题的背景下进行可伸缩的自适应网格划分。这一建议的优点将通过交付通用工程质量软件来展示，该软件能够在解决大规模流体-结构相互作用问题时利用数千个处理器，并且还将产生更好的方法和算法来转移到Petaflop性能。目前，Uintah软件已经在数千个处理器上进行了演示，这一提议的努力将确保它将使用容错和自适应算法继续在千万亿级机器上进行扩展。广泛影响：该提议的更广泛影响是，由此产生的算法、代码和所获得的性能水平将影响下一代代码开发人员，他们将寻求在通用代码上获得千万亿级的性能。尤其重要的是，可以使具有动态变化的工作模式的代码在大型CPU计算机器上执行。该项目的教育影响通过在犹他大学教授的高性能计算和计算科学项目来展示。可伸缩性和适当的算法和软件基础设施是研究生培训的一部分，作为科学计算多学科计划的一部分。

Intellectual Merit: Project proposes to improve and extend the Uintah framework to enable adaptive computations on modern HPC architectures with tens of thousands of nodes and complex communication/memory hierarchies. Uintah has been developed under the University of Utah Center for Simulation of Accidental Fires and Explosions (C-SAFE), a DOE-funded academic alliance project. This proposal aims to bring the unique capabilities that Uintah provides to the broader NSF science and engineering community. At the same time the underlying methodology employed will illustrate how such software packages may be used on large-scaleHPC architectures in a fault-tolerant and self-adaptive way. Uintah is applicable to a wide range of engineering domains that require fluid-structure interactions and highly deformable models, and is used in its present advanced beta-release form for NSF-funded bioengineeringand computer science projects at the University of Utah (such as tissue modeling), as well as other NIH, DARPA, DOE and industry funded projects both at Utah and other institutions (such as heart injury modeling and blast-wave simulation). The aim of this proposed project is to move this software out into the broader NSF computational science community. Uintah is open-source software available through the MIT open-source license. Uintah will be extended to provide additional fault tolerance through redundant computation, which will be facilitated by the component-based architecture of the Uintah system. We will add hierarchical Monte Carlo radiation transport models to Uintah to provide additional capabilities for heat transfer, and electromagnetic wave propagation. Finally, we will demonstrate scalability of Uintah on Petascale architectures for real-world engineering design problems and support external users of Uintah for NSF-funded and other research through modern collaborative tools such as an online wiki. There are two major expected outcomes from this project. The first direct outcome is an enhanced version of Uintah with the potential to scale to many thousands of processors and to begin to approach petaflop performance. This code will have immediate use in the computational modeling of many applications involving hazards. The second contribution is less direct but equally important. The use of the underlying task graph methodology inside Uintah will provide an instance of how a potentially important methodology may be applied to large-scale parallelism in a fault-tolerant and self-adaptive way. Other contributions as part of the above will include scalable adaptive meshing in the context of fluid-structure interaction problems.The merit of this proposal will be demonstrated through the delivery general-purpose engineering quality software capable of utilizing thousands of processors when solving large scale fluid-interaction problems, and will also result in better methodologies and algorithms for the move to petaflop performance. At present the Uintah software has been demonstrated on thousands of processors, and this proposed effort will ensure that it will continue to scale on petascale class machines using fault tolerance and self-adaptive algorithms.Broader Impact: The broader impact of this proposal is that the resulting algorithms, codes and levels of performance achieved will influence the next generation of code developers who will be seeking petascale performance on general purpose codes. Of particular importance is the idea that codes with dynamically varying work patterns can be made to perform on large cpu-count machines. The educational impact of this project demonstrated through the high-performance computing and computational science programs at taught at University of Utah. Scalability and the appropriate algorithms and software infrastructures are part of the graduate training that students receive as part of a multidisciplinaryprogram in scientific computing.