Symposium on Parallel solvers for very large PDE based systems in the Earth- and atmospheric sciences

Eike Mueller (University of Bath), Chris Maynard (Met Office), Graham Riley (University of Manchester)

Contact: e [dot] mueller [at] bath [dot] ac [dot] uk (subject: ParCo2015%20Mini%20Symposium) (e.mueller at



Lawrence Mitchell, Florian Rathgeber, and David A. Ham - Firedrake: automating the finite element method by composing abstractions
Daniel Ruprecht and Robert Speck - Fault-tolerance of the parallel-in-time integration method PFASST
M. Wittman, H. Stengel, G. Wellein, B. Gmeiner, U. Rüde, C. Waluga, B. Wohlmuth, M. Mohr, J. Weimüller, and H. P. Bunge - Performance Modeling and Analysis of Stencil operations in Earth Mantle Convection Simulations
A. R. Porter, R. W. Ford, M. Modani, M. Ashworth, and G. D. Riley - Towards Performance Portability in Earth-System Modelling with GOcean
Carlos Osuna, Tobias Gysi, Oliver Fuhrer, and Thomas C. Schulthess - Porting the COSMO dynamical core to heterogeneous platforms using STELLA Library
Andreas Dedner and Robert Klöfkorn - A parallel dynamical core based on the Discontinuous Galerkin method implemented in the open source package DUNE: Implementational details and comparison with the operational core COSMO


Virtually all areas in the Earth- and atmospheric sciences require the fast solution of very large systems of equations which arise from the discretisation of partial differential equations (PDEs). With ever increasing parallelism in modern HPC systems and the recent proliferation of different chip hardware (manycore CPUs, GPUs, Xeon Phis, ...) it becomes increasingly important to exploit all available concurrency in the algorithm and implement these solvers efficiently on massively parallel hardware. Furthermore, with the growing complexity of models it is crucial to use suitable abstractions in code development which allow the development of performance-portable and maintainable software.

This minisymposium will consider all these aspects by including presentations on state-of-the art solvers for systems with trillions (10^{12}) of unknowns in geophysical applications as well as the application of novel algorithms (such as parallel-in-time methods) and the efficient implementation of complex PDE solvers for advanced discretisations. In particular we will include talks on the GPU port of state-of-the art weather forecast models and on performance-portable tools based on automatic code generation, such as the python based PyOP2/firedrake toolchain. This and similar frameworks have recently been employed to successfully implement fast parallel solvers for finite-element based discretisations of the atmospheric equations of motion.


Extended Abstract submission deadline: June 30, 2015

Notification of acceptance: July 3, 2015

Full paper submission deadline: Jul 31, 2015

Mini Symposium: September 2, 2015


Last updated: 14 Aug 2015 at 15:06