Last updated: Aug 2018
IMGW:
0.6 FTE/COSMO year 2017 (Adam Ryczkowski, Marcin Polkowski, Zbigniew Piotrowski)
0.8 FTE/COSMO year 2018
1.2 FTE/COSMO year 2019
The goal of CEL-ACCEL is to implement the consistent anelastic/compressible COSMO-EULAG dynamical core (DC) within the C++ dynamical core framework of COSMO. The main deliverable is the high-resolution forecasting system ready for pre-operational testing on modern supercomputing architectures (GPU, manycore CPUs). Technical means include application of modern software engineering on accelerators, based on established COSMO-GPU strategies like Domain Specific Languages (DSL) and OpenACC.
The future very-high resolution weather forecasting of COSMO relies on both: robust dynamical cores allowing for accurate representation of convection interacting with steep mountain slopes, and computationally effective frameworks for integration of COSMO software. The former aspect is already addressed with the development of ICON-LAM and COSMO-EULAG dynamical cores, whereas the latter is realized with the GridTools/OpenACC port of COSMO/ICON software. In the broader context, substantial scientific progress is made within the PantaRhei (IFS-FVM) and ESCAPE H2020 projects led by ECMWF, benefiting the efforts of COSMO Consortium within the mentioned areas.
The newly funded “Numerical weather prediction for sustainable Europe” project, financed by the European Regional Development Fund within the Smart Growth Operational Programme 2014 - 2020, Measure 4.4. “Boosting human potential in R&D sector” through the First Team 1/2016 award of Foundation for Polish Science contributes to the aforementioned efforts. Thus it is justified to establish new COSMO priority project, complementary to the CELO priority project of COSMO, aiming at covering aspects that are not a part of the existing efforts but are of key importance to exploit COSMO-EULAG in operational production at COSMO weather services.
The goal of this priority project is to synthesize efforts at the leading edge of research and development in high-resolution numerical weather modelling. It aims at the development of highly efficient COSMO-EULAG port using modern software engineering standards and supercomputing architectures. The new formulation would exploit GridTools library in concert with the energy-efficiency technologies and implementations developed within the ESCAPE H2020 project. In particular, it should employ the concept of NWP dwarfs (mini-applications) that are ready for standalone verification and development by the independent research groups, yet can be easily reintegrated into COSMO framework.
Task 1 will be carried out from the beginning until the end of the project. Task 2 starts with the advent of COSMO year 2019.
Purpose of this task is to prepare COSMO-EULAG port capable to run on accelerators. The subtasks planned reflect current strategies of COSMO Consortium and include:
Deliverables:
(09.2017) Rewritten diffusion operator for COSMO-EULAG
(03.2018) Gridtools port of all COSMO-EULAG stencil components
(09.2018) COSMO-EULAG dynamical core in C++/Gridtools
(09.2019) Directive-based port of COSMO-EULAG
The ultimate goal of this priority project is to allow for operational forecast production on accelerators using COSMO-EULAG. Within this task, it is planned to perform stability and performance evaluation of COSMO-EULAG, as well as assessment of the verification scores, especially in the lower boundary layer. Appropriate subtasks are:
Deliverables:
(09.2019) Report of stability, performance and forecasting scores of COSMO-EULAG
Timely progress of this priority project relies on the timely progress of CELO priority project. The key CELO outcomes needed for the success of this project are: maintaining common (amongst CELO and CEL-ACCEL) code-base which includes the latest version of COSMO C++, OpenACC code base, robust and tested COSMO-EULAG on CPU, fully coupled to complete set of COSMO physics parameterizations and models (soil,etc.). For the realization of Task 5, deterministic COSMO-EULAG on CPUs should be already well tested, with long-term forecasting scores examined and at least partly tuned.
The realization of this project relies almost completely on the stable development of GridTools software.
Rojek K.A., M. Ciznicki, B. Rosa, P. Kopta, M. Kulczewski, K. Kurowski, Z.P. Piotrowski, L. Szustak, D. Wójcik, R. Wyrzykowski (2015), Adaptation of fluid model EULAG to graphics processing unit architecture, Concurr Comput., 27, 937-957