MRI: Acquisition of a High Performance Cluster for Multidisciplinary Computational Research

提案号：11-26438PI(s): deDoncker, Elise H.；Peter A. Guftafson；Kaugars Karlis;William W. Liou；机构：西密歇根大学职称：MRI/Acq。多学科研究项目的高性能集群提议：该项目，获得一个高性能计算机集群，支持以计算工程为重点的跨学科项目。这种混合性能计算集群能够模拟计算流体动力学、纳米管模型、柔性体空气动力学以及油墨转移和雕刻性能的纸张工程模型。计算集群的最初用途将包括计算机科学的基础研究，以及高能物理和医学物理项目的并行算法开发和实现，以及计算流体动力学（CFD）、碳纳米管（CNT）建模、机械和航空工程中的复合材料研究、柔性体空气动力学在造纸、化学工程和成像中的应用，以及化学中的新计算方法。由于混合计算环境支持节点之间的分布式内存计算和节点内的共享内存计算以及图形处理器核心上的大量并行计算，该集群将提供独特的资源。在多尺度建模中，为了对复合材料力学产生新的基本认识，需要大量的计算能力来处理尺度间的信息传递过程。对于高分辨率CFD， HPC计算机集群的提议适用于高速，并行DNS（直接数值模拟）和LES（大涡模拟）模拟（这在提高对大范围特征尺度下流体湍流的理解方面发挥关键作用）。计算机集群有望减少计算含有碳纳米管的复杂结构的力学行为的时间，其规模可与实验室环境中的生产相媲美。传统的CFD方法无法准确地模拟与三维流动高度相互作用的柔性结构，而提出的Lattice Boltzman方法为复杂的柔性体空气动力学提供了强大的计算工具。高性能计算化学有望对还原异构酶（DXR）催化的反应机理有新的认识，并通过结合价键和分子轨道理论，开发新的方法来模拟化学键的性质。在计算机科学方面，将开发一种新的并行积分模型来支持迭代积分的自动计算，使新的费曼环积分计算能够支持在对撞机上观察到的基本粒子碰撞的理论物理模型的验证。迭代积分还将应用于医学物理计算中，以调节不同角度的质子束辐射剂量。更广泛的影响：该仪器为四个系和两个学院的尖端计算设施提供了培训机会。它允许学生参与系统的操作和维护。新的集群应该使促进跨学科研究和合作的项目成为可能。费曼集成项目由位于日本筑波的高能加速器研究组织支持。放射治疗项目的合作是与底特律亨利福特医院的工作人员建立的。此外，一些项目还应用于生物医学工程（复合材料建模、骨科植入物和其他医疗装置、纳米管给药、静脉注射管）、医学物理学（放射肿瘤学）和生物化学（抗生素、抗疟疾药物和除草剂的开发），这可以作为生物工程框架内合作和计划中的大学医学院的基础。复合材料的多尺度建模研究可能会通过提高燃料经济性和减轻结构部件的重量来影响全球能源使用。它还应通过风力涡轮机模型和其他可再生能源发电技术的进步，为改善环境作出贡献。柔性体空气动力学的研究结果将更新、推进并积极影响我们对柔性和可变形纤维和生物颗粒的迁移、旋转、絮凝和分散的基本机制的理解。因此，这些仪器可以为这个主要为本科生服务的机构的教育发展奠定基础。还将传播项目的软件、数据和辅助材料。

Proposal #: 11-26438PI(s): deDoncker, Elise H.; Guftafson, Peter A.; Kaugars, Karlis; Liou, William W.; Mo, YirongInstitution: Western Michigan University Title: MRI/Acq.: High Performance Cluster for Multidisciplinary Research Project Proposed:This project, acquiring a high performance computer cluster, supports interdisciplinary projects focused on computational engineering. This hybrid performance computing cluster enables simulations of computational fluid dynamics, nanotube models, flexible body aerodynamics, and paper engineering models of ink transfer and engraving properties. The initial uses of the computational cluster will include basic research in Computer Science, with parallel algorithm development and implementation for projects in high energy physics and medical physics, and furthermore in computational fluid dynamics (CFD), carbon nanotube (CNT) modeling, and composite materials research in Mechanical and Aeronautical Engineering, flexible body aerodynamics in Paper, Chemical Engineering, and Imaging, and novel computational approaches in Chemistry. With a hybrid computing environment supporting distributed memory computations between nodes and shared memory computation within nodes and massive parallel computation on graphics processor cores, this cluster would provide a unique resource.Significant computational capability is required for the process of inter-scale information transfer in multi-scale modeling to yield new fundamental understanding of composite mechanics. For high-resolution CFD, the proposed acquisition of the HPC computer cluster is suited for high-speed, parallel DNS (Direct Numerical Simulations) and LES (Large Eddy Simulations) simulations (which play a key role in the advancement of the understanding of fluid turbulence in a wide range of characteristic scales). The computer cluster is expected to reduce the wall-clock time for computing the mechanical behavior of complex structures containing CNTs at a scale comparable to their production in a laboratory environment. While conventional CFD methods fail to accurately model flexible structures highly interacting with 3D flow, the proposed Lattice Boltzman method provides a powerful computational tool for complicated flexible body aerodynamics. High performance computational chemistry is expected to contribute new understanding of the reaction mechanisms catalyzed by DXR (reductoisomerase) and develop new methods for modeling the nature of chemical bonding by combining valence bond and molecular orbital theories. In computer science, a new model of parallel integration will be developed to support the automatic computation of iterated integrals, enabling new computations of Feynman loop integrals to support the validation of theoretical physics models for the collision of elementary particles observed at colliders. Iterated integration will also be applied in medical physics computations for regulating the dosage of proton beam radiation at varied angles. Broader Impacts: This instrumentation provides the opportunity for training on cutting-edge computing facilities across four departments and two colleges. It allows student participation in the operation and maintenance of the system. The new cluster should enable projects that foster inter-disciplinary research and collaborations. The Feynman integration project is supported by the High Energy Accelerator Research Organization in Tsukuba, Japan. Collaboration for the radiation therapy project is established with staff of the Henry Ford hospital in Detroit. Furthermore, some projects have applications in biomedical engineering (composite materials modeling, orthopedic implants and other medical devices, drug delivery with nanotubes, flow in IV tubes), medical physics (radiation oncology) and biochemistry (development of antibiotics, antimalarial drugs and herbicides), which can underlie cooperation in the framework of bio-engineering and the planned medical school of the university. The research in multiscale modeling for composite materials might have impact on global energy usage through improved fuel economy associated with weight reduction in structural components. It should also contribute to improve the environment through advances in wind turbine models and other renewable energy generating technologies. The results in flexible body aerodynamics should update, advance, and positively impact our understanding of the fundamental mechanisms of migration, rotation, flocculation, and dispersion of flexible and deformable fibers and bio-particles. Hence, the instrumentation can set the stage for educational developments at this mainly undergraduate serving institution. Software, data, and supporting materials of the projects will also be disseminated.