Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
7156954 | Computers & Fluids | 2015 | 39 Pages |
Abstract
Computational fluid dynamics (CFD) simulations generally require massive computing power. Supercomputers are therefore typically adopted for these tasks. Recently, the Hadoop platform for data-intensive and distributed computing was introduced with a programming model called MapReduce. Hadoop offers several benefits, including automatic parallelizing/distributing and high availability, without requiring expensive hardware. In this paper, we propose an approach to developing a computational fluid dynamics simulation with a finite-volume method based on the Hadoop platform and inexpensive hardware. Our approach employs OpenCL to enable heterogeneous machine optimization and to control the general-purpose graphics processing unit. As a case study, we implement a system for magnetohydrodynamics (MHD) simulation that is an essential part of CFD simulations. We describe the design of our MHD simulator and present the experimental results. The results show that our approach outperforms the conventional solutions.
Keywords
Related Topics
Physical Sciences and Engineering
Engineering
Computational Mechanics
Authors
Milhan Kim, Youngjun Lee, Ho-Hyun Park, Sang June Hahn, Chan-Gun Lee,