| |
 |
| Paper: |
New Developments in Modeling MHD Systems on High Performance Computing Architectures |
| Volume: |
406, Numerical Modeling of Space Plasma Flows: ASTRONUM - 2008 |
| Page: |
223 |
| Authors: |
Germaschewski, K.; Raeder, J.; Larson, D.J.; Bhattacharjee, A. |
| Abstract: |
Modeling the wide range of time and length scales present even in fluid models of plasmas like MHD and X-MHD (Extended MHD including two fluid effects like Hall term, electron inertia, electron pressure gradient) is challenging even on state-of-the-art supercomputers. In the last years, HPC capacity has continued to grow exponentially, but at the expense of making the computer systems more and more difficult to program in order to get maximum performance. In this paper, we will present a new approach to managing the complexity caused by the need to write efficient codes: Separating the numerical description of the problem, in our case a discretized right hand side (r.h.s.), from the actual implementation of efficiently evaluating it. An automatic code generator is used to describe the r.h.s. in a quasi-symbolic form while leaving the translation into efficient and parallelized code to a computer program itself. We implemented this approach for OpenGGCM (Open General Geospace Circulation Model), a model of the Earth’s magnetosphere, which was accelerated by a factor of three on regular x86 architecture and a factor of 25 on the Cell BE architecture (commonly known for its deployment in Sony’s PlayStation 3). |
|
|
|
|
|
