Massively parallel direction splitting algorithms for complex flow

复杂流的大规模并行方向分裂算法

• 批准号: 216926-2012
• 负责人: Minev, Peter
• 金额: $ 1.89万
• 依托单位: University of Alberta
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2015
• 资助国家: 加拿大
• 起止时间: 2015-01-01 至 2016-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=570064
• 关键词: Massively; parallel; direction; splitting; algorithms

The current trend in the development of High Performance Computers (HPC) is to create distributed clusters of very large number of cores (the next generation of HPC are expected to have of the order of one million cores). Therefore, one of the most important features of simulation algorithms for scientific computing is their performance when implemented on large parallel clusters. The classical algorithms for discretization of Partial Differential Equations (PDEs) are not always well suited for such a task. Recently we have developed a very fast and parallelizable direction-splitting algorithm for incompressible and the so-called variable density incompressible flows. It enabled us to perform simulations involving billions of unknowns and placed us at the forefront of large scale parallel computational fluid dynamics. The goals of the present project are: (i) To extend and analyze the algorithm to flows with rigid particles and perform a very large scale simulation of such flows. The hydrodynamic interaction of millions of particles is certainly within the reach of these methods. Then we will use the code to study several flows of interest in engineering applications. (ii) Develop the method for fluid-structure interaction problems and use it for simulation of various physiological flows which require a very large number of degrees of freedom. (iii) Start work on the development and analysis of direction splitting methods on the sphere. This part of the project is the initial stage of a longer-term goal to develop a simulation tool for atmospheric and oceanic flows. Since the method is extremely fast and parallelizable, we believe that it will allow us to perform simulations on the scale of the entire globe of the coupled atmospheric and ocean flow.

目前高性能计算机（HPC）的发展趋势是创建具有大量核心的分布式集群（下一代HPC预计将拥有100万个核心）。因此，科学计算仿真算法最重要的特征之一是它们在大型并行集群上实现时的性能。经典的偏微分方程离散化算法并不总是适合于这样的任务。最近，我们针对不可压缩流和所谓的变密度不可压缩流开发了一种非常快速且可并行化的方向分裂算法。它使我们能够执行涉及数十亿未知数的模拟，并将我们置于大规模并行计算流体动力学的前沿。