|
Titel |
Massive parallel simulation of large scale subsurface flow |
VerfasserIn |
Olaf Ippisch, Markus Blatt, Peter Bastian |
Konferenz |
EGU General Assembly 2011
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 13 (2011) |
Datensatznummer |
250051622
|
|
|
|
Zusammenfassung |
Simulation of subsurface flow is crucial for the prediction and control of groundwater
production, the assessment of water contamination and becomes more and more important for
flood and climate prediction.
We implemented the highly scalable simulation program μφ (muPhi) for the solution of
Richards’ equation. The code is capable to describe saturated and unsaturated subsurface
flow. A cell-centred Finite-Volume scheme is used for the spatial discretisation, an implicit
Euler scheme for the time discretisation and an incomplete Newton-scheme for the
linearisation of the non-linear equations. The linear equation system is solved with the
iterative solver template library (ISTL) developed by our group in the framework of the
DUNE-Project. A CG or BiCGstab solver with an algebraic multigrid preconditioner
allows the efficient and scalable solution of huge systems on massively parallel
computers.
We tested the simulation code with realistic three-dimensional saturated and unsaturated
subsurface flow problems on Europe’s fasted supercomputer JUGENE and also with a
48-core off-the-shelf server. The results demonstrate excellent scalability up to JUGENE’s
294’912 cores. A groundwater flow problem with an anisotropic autocorrelated random
permeability field with a range of 9 orders of magnitude and 150 billion unknowns can be
solved in less than 15 minutes. This allows the solution of large scale subsurface flow
problems with high spatial resolution. |
|
|
|
|