Parallel Computing in Geosciences
UPPMAX, project No. P2004009, 2004 - 2006
Contents: Description Aims Solved problems Used computers Reports Team Events
Description
The project represents multi-disciplinary research involving mathematical modelling of problems arising in geophysics and geo-engineering applications, numerical solution methods with an emphasis on preconditioned iterative techniques and their efficient implementation on parallel computers. The geosciences and geo-engineering nowadays consider a variety of problems, which demand the use of sophisticated large-scale computer modelling. The target problems with this project include modelling of stress field changes due to mining and underground constructions, assessment of underground nuclear waste repository projects as well as elastic and viscoelastic effects of the Earth due to other glacial advance and recession.
The very high complexity of these problems is usually determined by the required models. As an example, the mechanical behaviour of rocks can be described by models ranging from linear elasticity to models involving nonlinear material response, permanent deformation, rheology phenomena or influence of thermal changes and flow in porous media. The large-scale arises when we consider large 3D domains to enable monitoring of the far field effects as well as to be able to specify realistic boundary conditions. Simultaneously, some engineering decisions are based on a detailed analysis of the near field processes.
The solution of the above described problems then requires the use of robust and numerically efficient iterative solution methods and their implementation on modern parallel computer architectures. This project addresses both discretization of the mathematical models by the finite element method and solution of the discretized problems by robust and efficient iterative methods using domain decomposition ideas combined with hierarchical substructuring approaches.
The considered numerical methods include Krylov subspace methods preconditioned by Schwarz-type decomposition methods accelerated by an auxiliary coarse grid problem. Such methods are applied to a hierarchy of problems including elasticity, thermo-elasticity, visco-elasticity and poro-elasticity.
Aims of project
- To implement the developed methods on various types of computer architectures using different programming paradigms (MPI and OpenMP) and to improve the overall parallel performance of the methods to insure short simulation times.
- To compare the scalability and performance of the solvers on various computer platforms.
Solved problems
-
DR
- development of stresses due to mining.
Large-scale 3D geotechnical model, Dolni Rozinka, the Czech Republic. -
KBS
- modelling thermo-mechanical phenomena.
Large-scale 3D geoenvironmental model, Aspo Prototype Repository, Sweden.
Used computers
- Lomond - Sun HPC cluster, EPCC, Edinburgh
- Natan - IBM symmetric multiprocessor, IGAS, Ostrava
- Thea - Beowulf cluster, IGAS, Ostrava
- Termit - Beowulf cluster, VSB-TU, Ostrava
- Simba - Sun symmetric multiprocessor, UPPMAX, Uppsala
- Ra - Sun cluster, UPPMAX, Uppsala
Reports and presentations
- R. Blaheta: Modelling of thermo-mechanical phenomena arising from the underground disposal of the spent nuclear fuel. Uppsala, November 23, 2004
- J. Stary: Application of parallel computing to elasticity and thermoelasticity problems. Uppsala, November 23, 2004
- G. Bencheva: MPI implementation of a PCG solver for nonconforming FEM problems: overlapping of communications and computations. Uppsala, May 12, 2005
- G. Bencheva, S. Margenov, J. Stary: Parallel PCG Solver for Nonconforming FEM problems: Overlapping of Communication and Computations. LSSC 2005, Springer, LNCS. Conference Large-Scale Scientific Computations 2005, Sozopol, June 6 - 10, 2005
- R. Blaheta, R. Kohut, M. Neytcheva, J. Stary: Schwarz Methods for Discrete Elliptic and Parabolic problems with an Application to Nuclear Waste Repository Modelling. Submitted to Mathematics and Computers in Simulation, IMACS/Elsevier, special issue Modelling 2005. IMACS conference Modelling 2005, Pilsen, July 4 - 8, 2005
Research team
- UU - Scientific Computing Group, Department of Information Technology, Uppsala University, Sweden
- IGAS - Department of Applied Mathematics and Computer Science, Institute of Geonics, Academy of Sciences of the Czech Republic, Ostrava, the Czech Republic
- IPP - Scientific Computing Department, Institute for Parallel Processing, Bulgarian Academy of Sciences, Sofia, Bulgaria
- ARTEC - Research Centre for Advanced Remediation Technologies, Technical University of Liberec, the Czech Republic
Events
- Workshop on Modelling and numerical simulation methods for large-scale problems in Geophysics and Geotechnics. Robustness, numerical efficiency, scalability. Uppsala, November 23, 2004
- Workshop on Numerical Linear Algebra Methods for Partial Differential Equations. Uppsala, May 12, 2005
- Workshop on Numerical analysis: iterative methods in modelling and simulation of challenging engineering problems. Ostrava, June 28, 2005
- Workshop on Numerical analysis: iterative methods in modelling and simulation of challenging engineering problems II. Ostrava, November 2, 2006