T.I Gombosi, K.G. Powell,
Q F. Stout,
E.S. Davidson, D.L. DeZeeuw, L.A. Fisk, C.P.T. Groth, T.J. Linde, H.G.
Marshall, P.L. Roe, B. van Leer
Center for Space Environment Modeling (CSEM)
University of Michigan
Abstract: A new magnetohyrdrodynamics (MHD) code has been developed for massively parallel computers, using adaptive mesh refinement (AMR) and a new 8-way Riemann solver. A new data structure, adaptive blocks, has been developed to help sustain high performance on AMR. The code was implemented on a Cray T3D parallel computer with 512 processors. In the first application, which modeled the expansion of the solar wind from the solar surface, the code sustained 13 GFLOPS. This code is the product of a NASA Grand Challenge Investigor Team, and this is merely the initial results of an on-going project.
Much of our work has also been applied to fluid dynamics problems. In particular, our combined use of Roe solvers and AMR had proven quite useful, and the introduction of adaptive blocks has greatly increased achieved performance. (Here is a paper about adaptive blocks).
Keywords: heliosphere, multi-scale model, space plasma, parallel computing, supercomputing, Cartesian adaptive grid
Complete paper: postscript pdf This paper appears in High Performance Computing 1997, pp. 46-51
|Copyright © 2005-2016 Quentin F. Stout|