Parallel solution of ocean circulation flow problems
The study of ocean flow patterns is important for the understanding of the climate. By this study one hopes to be able to predict long-term regional climate changes which may entail extreme events like flooding. This will allow decision makers to take adequate measures in time.
At the Institute for Marine and Atmospheric Research Utrecht (IMAU) a package THCM has been eveloped by which such studies can be performed. In THCM large systems of equations need to be solved frequently. This activity often takes the major part of the total computation time (over 90% is not unusual). Therefore parallelisation of this part is essential. On sequential computers MRILU is used to solve these systems. MRILU (Matrix Renumbering Incomplete LU) is a linear system solver for arbitrary sparse matrices arising in engineering. It is developed at the mathematics department of the University of Groningen.
For the MPI-based parallelisation of MRILU for an SGI Origin3800, the algebraic domain decomposition idea is used. On the subdomains MRILU is employed and ILUT is used for the system arising from the interfaces. These result in a global preconditioner which may differ from iteration step to iteration step. Therefore, FGMRES is used to solve the global preconditioned system. A more detailed explanation can be found in the Technical Description and the Users Guide.
For the partitioning, METIS-4.0 is used. Furthermore one needs the MRILU library and a special shell around it. We emphasize that the shell is dedicated to the ocean circulation flow problem. Hence, for other applications the shell needs to be adapted.
For implementational details contact A.Meijster. For details on the solution process contact F.W.Wubs.
This development was sponsored by the national foundation for computerfacilities NCF
|Laatst gewijzigd:||24 november 2015 10:47|