Convective Turbulence and Mixing in Astrophysics

Objective: To develop a new generation of highly portable production codes for hydrodynamics and magneto-hydrodynamics [MHD] relevant to astro-physics, which take advantage of the large memory and fast computational speeds of modern parallel machines.

Approach: We are using tools provided by Argonne National Laboratory (ANL) to develop portable codes for distributed memory environments; are working with ANL computer scientists to refine these tools; are using Univ. of Colorado (CU) performance tools to evaluate and improve parallel code efficiency, and to study and improve network communications (including ATM); and are comparing parallel computing strategies by also investigating alternative parallelization tools based on shared memory constructs.

Accomplishments: (1) Developed new application codes for MIMD architectures, including: (a) a code for 3-D compressible hydro-dynamics, with non-linear thermal conduction, based on a higher-order Godunov method, and a nonlinear multigrid elliptic solver; (b) a 3-D fully parallel fast fourier transform, for spectral or pseudospectral hydrodynamics and mag-netohydrodynamics [MHD] codes; (c) a compressible hybrid code, which performs spectral transforms on individual processors, and finite difference code across processors (for convection in rotating systems); Machines used in the development of the above codes include the IBM SP-1, the Intel Delta and Paragon, the KSR, and the Cray Y-MP. In addition, we have (d) developed an MHD code, using FORTRAN 90, on the CM5 (which also runs on the Cray C90, with changes in I/O and in one initialization statement only); (e) collaborated with GSFC and JPL scientists on the development of new hydrodynamic codes on MPPs.

(2) Used our new codes for novel scientific calculations, including: (a) compressible penetrative convection in Sun-like stars; (b) core convection within A stars; (c) compressible turbulent convection constrained by rotation; (d) evolution of magnetic fields in turbulent conducting fluids; (e) compressible turbulent magnetoconvection, and subgrid modeling for convection; (3) Studied the performance of parallel machines and communications, including: (a) considerable testing of ATM networks; (b) testing and refinement of the PETSc library for scientific computing.

Significance: We have developed the first generation of fully portable application codes for turbulent mixing problems on massively parallel architectures, representing a broad range of techniques for solving hydrodynamic problems. These codes achieve the high performance required to study new frontier problems in astrophysical fluid dynamics.

Status/Plans: Our plans include improving: (a) the I/O capabilities of available parallel machines, and (b) the efficiency of codes running on individual processors.

For example, with the aid of CU performance monitoring tools, we aim to improve the singleprocessor performance of our codes (especially in their utilization of caches). We will implement fully parallel I/O, in collaboration with ANL. Finally, we will mount a major effort in data compression, with the aim of reducing the I/O bandwidth demands of high-performance fluid dynamic codes; and will expand the types of machines used to include the Cray T3D, and Convex Exemplar, and a cluster of DEC Alphas.

FIGURE CAPTION:

Temperature fluctuations [dark: cool; light: hot] in a layer of convectively unstable gas (upper half) overlying a convectively stable layer (lower half) within the deep interior of a Sun-like star. This simulation was carried out on the ANL IBM SP-1.

Principal Investigator Progress Metric(s)

Point of Contact:

Return to the PREVIOUS PAGE

curator: Larry Picha (lpicha@cesdis.gsfc.nasa.gov)