![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Geodynamic modelling using finite difference and finite element methods on massive parallel computers. |
VerfasserIn |
Boris Kaus, Dave May |
Konferenz |
EGU General Assembly 2011
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 13 (2011) |
Datensatznummer |
250054057
|
|
|
|
Zusammenfassung |
3D geodynamic modelling of lithospheric-scale deformation remains a challenge as it
requires scalable multilevel preconditioners which perform well in the presence of
large and abrupt jumps in viscosity. In recent years, we have developed LaMEM
(Lithosphere and Mantle Evolution Model), which is a finite element code, that can
be combined with a marker-and-cell method (to model large strains) and can be
used in an Eulerian, Lagrangian or ALE manner. LaMEM is built on top of the
parallel PETSC package (http://www.mcs.anl.gov/petsc), which allows us to access
a large repertoire of smoothers, iterative solvers, as well as algebraic multigrid
preconditioners such as ML. We can use either Q1P0, stabilized Q1Q1 or Q2P-1 elements (all
configurable from the command-line). Recently, we have also added a staggered grid finite
difference discretisation. Here, we will compare the relative performance of the
FD versus the FEM method for a few selected benchmark cases. In addition, we
have ported LaMEM to massive parallel machines (Cray XT5 and IBM Bluegene).
Using up to 2048 cores, we obtain ~ 70% efficiency in weak scaling tests, thereby
enabling problems with more than 500 million degrees of freedom to be solved. |
|
|
|
|
|