Monday, June 1, 2015 |
HG E1.1 |
| 13:00 - 13:30 |
An Accurate and Efficient Numerical Framework for Adaptive Numerical Weather Prediction, Giovanni Tumolo (The Abdus Salam ICTP, Italy) Abstract |
Monday, June 1, 2015
HG E1.1, 13:00 - 13:30
An Accurate and Efficient Numerical Framework for Adaptive Numerical Weather Prediction; Giovanni Tumolo (The Abdus Salam ICTP, Italy)
Co-Authors: Luca Bonaventura (Politecnico di Milano, Italy)
We present an adaptive discretization approach for model equations typical of NWP, which combines the semi-Lagrangian technique with a TR-BDF2 semi-implicit time discretization and with a DG spatial discretization with (arbitrarily high) variable and dynamically adaptive element degree. The resulting method has full second order accuracy in time, is unconditionally stable and can effectively adapt at runtime the number of degrees of freedom employed in each element, in order to balance accuracy and computational cost. Numerical results of classical 2D benchmarks for shallow water equations on the sphere and Euler equations on a vertical slice confirm the potential of the proposed formulation.
| 13:30 - 14:00 |
Towards a 3D Dynamical Core Based on Mixed Finite Element Methods, Andrew McRae (Imperial College London, United Kingdom) Abstract |
Monday, June 1, 2015
HG E1.1, 13:30 - 14:00
Towards a 3D Dynamical Core Based on Mixed Finite Element Methods; Andrew McRae (Imperial College London, United Kingdom)
Co-Authors: Colin Cotter (Imperial College London, United Kingdom); David Ham (Imperial College London, United Kingdom)
There are known scaling issues with the otherwise-successful combination of C-grid finite difference methods on a global latitude-longitude grid. Direct extensions of C-grid methods to quasi-uniform grids have known inadequacies. Mixed finite element methods, recently unified under the label of finite element exterior calculus, provide a way of recovering desirable numerical properties but on arbitrary meshes. After some initial work in 2D, we implemented a method for the automated generation of sophisticated 'tensor-product' finite elements. The use of finite element methods also invites the possibility of high-order approaches, though we are not actively exploring this.
| 14:00 - 14:30 |
Scalable Dynamic Load Balancing of Detailed Cloud Physics with FD4, Matthias Lieber (Technische Universität Dresden, Germany) Abstract |
Monday, June 1, 2015
HG E1.1, 14:00 - 14:30
Scalable Dynamic Load Balancing of Detailed Cloud Physics with FD4; Matthias Lieber (Technische Universität Dresden, Germany)
Co-Authors:
Load balancing of large-scale scientific simulations is a challenging task. My talk shows how the open source framework FD4 (Four-Dimensional Distributed Dynamic Data structures) enables scalable dynamic load balancing of detailed cloud physics in the atmospheric model COSMO-SPECS+FD4. The concept is based on separating data structures and decomposition such that the cloud physics model can be partitioned independently of the atmospheric model. To increase scalability, a hierarchical space-filling curve partitioning algorithm is used. Benchmarks on a Blue Gene/Q system show the scalability up to 256k ranks. FD4 can also be applied to other multiphase or multiphysics problems.
| 14:30 - 15:00 |
Using GPUs Productively for the ICON Climate Model, William Sawyer (CSCS / ETH Zurich, Switzerland) Abstract |
Monday, June 1, 2015
HG E1.1, 14:30 - 15:00
Using GPUs Productively for the ICON Climate Model; William Sawyer (CSCS / ETH Zurich, Switzerland)
Co-Authors: Markus Wetzstein (CSCS / ETH Zurich, Switzerland); Leonidas Linardakis (Max Planck Institute for Meteorology, Germany)
The Swiss National Supercomputing Centre (CSCS) has undertaken a full port of the Icosahedral Non-hydrostatic (ICON) climate model currently under development at the Max Planck Institute for Meteorology (MPI-M) and the German Weather Service (DWD). Central to this port is the co-design with MPI-M and DWD developers to ensure that the code changes are minimal and non-intrusive, do not impose any significant changes in CPU performance, and can be incorporated directly into the development trunk. In this talk we present some of the techniques which facilitate this co-design project and initial results indicating that the performance is directly related to the peak memory bandwidth.
| 15:30 - 16:00 |
Development of a Vertical Slice Model Using Mixed FEM Discretizations, Hiroe Yamazaki (Imperial College London, United Kingdom) Abstract |
Monday, June 1, 2015
HG E1.1, 15:30 - 16:00
Development of a Vertical Slice Model Using Mixed FEM Discretizations; Hiroe Yamazaki (Imperial College London, United Kingdom)
Co-Authors: Colin Cotter (Imperial College London, United Kingdom); Jemma Shipton (Imperial College London, United Kingdom); Lawrence Mitchell (Imperial College London, United Kingdom); David A Ham (Imperial College London, United Kingdom); Andrew McRae (Imperial College London, United Kingdom)
A vertical slice model is developed for the Euler-Boussinesq equations with a constant temperature gradient in the y-direction (the Eady-Boussinesq model). The model is a solution of the full 3D equations with no variation normal to the slice, which is an idealized problem used to study the formation and subsequent evolution of weather fronts. Mixed FEM discretizations on an extruded mesh are performed using Firedrake, a high-level Python framework for the portable solution of PDEs on unstructured meshes. I will discuss the details of the vertical slice model implementations and the preliminary results obtained through test simulations developed to implement the modelling of fronts.
| 16:00 - 16:30 |
Nonlinear Stabilization Techniques for Finite Element Approximations of Fluid Problems, Murtazo Nazarov (Uppsala University, Sweden) Abstract |
Monday, June 1, 2015
HG E1.1, 16:00 - 16:30
Nonlinear Stabilization Techniques for Finite Element Approximations of Fluid Problems; Murtazo Nazarov (Uppsala University, Sweden)
Co-Authors:
This talk will discuss recent developments in stabilized finite element approximations using nonlinear viscosity methods. The regularization terms are constructed by the residual of the system or entropy equations. The method is successfully applied to compressible and incompressible Navier-Stokes equations and variable density flows in two and three space dimensions. Then we discuss about maximum principle preserving continuous finite element schemes for scalar conservation laws. The new method does not require any a priori knowledge of quantities like local wave-speed, proportionality constant, or local mesh-size and it is independent of the cell type.
| 16:30 - 17:00 |
preCICE - a Library for Flexible Surface-Coupling on Massively Parallel Systems, Benjamin Uekermann (Technische Universität München, Germany) Abstract |
Monday, June 1, 2015
HG E1.1, 16:30 - 17:00
preCICE - a Library for Flexible Surface-Coupling on Massively Parallel Systems; Benjamin Uekermann (Technische Universität München, Germany)
Co-Authors: Hans-Joachim Bungartz (Technische Universität München, Germany); Florian Lindner (Universität Stuttgart, Germany); Miriam Mehl (Universität Stuttgart, Germany)
With increasing compute power, the simulation of multi-physics gains more and more popularity. At the same time, a flexible software development process including existing single-physics codes becomes a necessity to cope with the overall complexity. preCICE is an open-source library to surface-couple single-physics codes. Its high-level API allows to write adapters for single-physics codes in 30 lines. Afterwards various codes can be coupled in a nearly plug-and-play manner. preCICE offers methods for interpolation, means for parallel communication, and sophisticated fix-point acceleration schemes. In this talk, we present our efforts and tests on the parallel efficiency of preCICE.
| 17:00 - 17:30 |
Preconditioned Variational Multiscale Stabilization (PVMS) for Low Mach Flows, Margarida Moragues Ginard (Barcelona Supercomputing Center, Spain) Abstract |
Monday, June 1, 2015
HG E1.1, 17:00 - 17:30
Preconditioned Variational Multiscale Stabilization (PVMS) for Low Mach Flows; Margarida Moragues Ginard (Barcelona Supercomputing Center, Spain)
Co-Authors: Guillaume Houzeaux (Barcelona Supercomputing Center, Spain); Mariano Vázquez (Barcelona Supercomputing Center, Spain)
In this work we present an implicit scheme for solving the Navier-Stokes equations at Low Mach regimes. The method can cope efficiently with both stiffness and numerical instabilities thanks to the following procedure. A local preconditioner is applied to the Navier-Stokes equations. Next, the preconditioned local system is discretized according to the Variational Multiscale Stabilization (VMS) method. Finally, a monolithic implicit Jacobi (i.e. fixed point) scheme is applied, which, in turn, uses an algebraic diagonal preconditioner on a GMRES iterative solver. The scheme is assessed on large-scale 3D problems, particularly on its accuracy, convergence and parallel efficiency.
| |
Tuesday, June 2, 2015 |
HG E1.1 |
| 10:30 - 11:00 |
High-Performance Computing and Big Data Challenges for NWP and Climate, Peter Bauer (ECMWF, United Kingdom) Abstract |
Tuesday, June 2, 2015
HG E1.1, 10:30 - 11:00
High-Performance Computing and Big Data Challenges for NWP and Climate; Peter Bauer (ECMWF, United Kingdom)
Co-Authors:
There is growing concern that the energy cost to produce weather forecasts, to quantify their uncertainty, and to handle the associated I/O load from enhanced observing systems as well as large forecast ensembles will be unaffordable with current HPC and data management technologies. NWP centres face enormous challenges due to the rising cost of energy associated with running complex high-resolution models on more and more processors in time for delivering forecasts with substantial societal benefit.
| 11:00 - 11:30 |
High-Order Schemes for Schemes for Atmospheric Dynamics on Unstructured Meshes, Antonios Antoniadis (Cranfield University, United Kingdom) Abstract |
Tuesday, June 2, 2015
HG E1.1, 11:00 - 11:30
High-Order Schemes for Schemes for Atmospheric Dynamics on Unstructured Meshes; Antonios Antoniadis (Cranfield University, United Kingdom)
Co-Authors: Panagiotis Tsoutsanis (Cranfield University, United Kingdom)
This paper presents an extension of the Weighted Essentially Non-Oscillatory (WENO) type schemes for the non-hydrostatic compressible Euler equations in conjunction with two and three-dimensional unstructured meshes. The schemes are suitable for regional and global climate models dynamical cores. Their potential lies in their simplicity; accuracy; robustness; non-oscillatory properties; versatility in handling any type of grid topology; computational and parallel efficiency. The WENO schemes (up to 5th -order) are applied to two- and three-dimensional test cases: a 2D rising thermal bubble; the 2D density current and the 3D Robert smooth bubble.
| 11:30 - 12:00 |
2D Adaptivity for 2.5D Flow Problems, Michael Bader (Technische Universität München, Germany) Abstract |
Tuesday, June 2, 2015
HG E1.1, 11:30 - 12:00
2D Adaptivity for 2.5D Flow Problems; Michael Bader (Technische Universität München, Germany)
Co-Authors: Kaveh Rahnema (Technische Universität München, Germany); Oliver Meister (Technische Universität München, Germany)
Many geophysical fluid dynamics problems feature a noticeably smaller extension in the vertical dimension compared to the horizontal dimensions. For such scenarios, we evaluate the potential of 2.5D grids that extend dynamically adaptive triangular meshes by a
uniformly refined third dimension. While we tolerate a slight increase in the total number of degrees of freedom, we aim at improved performance due to vectorisation along the vertical dimension. Examples will be presented for simulation of tsunamis and porous media flow.
| 12:00 - 12:30 |
A Conservative and Grid Adaptive Stabilization Scheme for Spectral Elements Based on a Dynamic SGS Model for LES. Application in Numerical Weather Prediction, Simone Marras (Naval Postgraduate School, USA) Abstract |
Tuesday, June 2, 2015
HG E1.1, 12:00 - 12:30
A Conservative and Grid Adaptive Stabilization Scheme for Spectral Elements Based on a Dynamic SGS Model for LES. Application in Numerical Weather Prediction; Simone Marras (Naval Postgraduate School, USA)
Co-Authors: Francis X. Giraldo (Naval Postgraduate School, USA)
The solution of the Euler equations by high-order spectral elements is prone to instabilities that must be damped in some way. We approach the problem of stabilization via an adaptive subgrid-scale scheme meant to treat the instabilities by modeling the sub-grid scale features of the flow. The equations are regularized via a dynamically adaptive stress proportional to the residual of the unperturbed equations. Its effect is close to none where the solution is smooth and it increases elsewhere. This is a first step toward LES for hurricanes and extreme weather forecast with the Nonhydrostatic Unified Model of the Atmosphere, NUMA, the dynamical core of the next-generation Navy model NEPTUNE.
