Implementation of ICON global 1km atmosphere-only demonstrator and performance analysis (D2.9)

Abstract

The ICON framework has been considered for the use in global high-resolution, atmosphere-only weather and climate simulations. Focus was put on two scenarios:
• an aqua-planet experiment using ECHAM physics and
• a realistic full-world simulation scenario in the context of numerical weather prediction.
Both scenarios were investigated at various grid resolutions and compute settings. The experiments—based on the partition compute2 of supercomputer Mistral—suggest that ICON is expected to deliver good scalability for global high-resolution simulations up to 1.2km when filling up to 70-100 horizontal grid cells (depending on the number of vertical grid levels used) per compute core. The main bottlenecks of the “computational workflow” for high-resolution simulations were found in the creation of initial data and external parameters for the large-scale runs, as well as in writing output; the latter is due to processing large amounts of data (200GB per time slice in a 2.5km globally resolved simulation) and is subject to reimplementation and improvement with regard to using asynchronous I/O on distributed memory systems.

About this document

• Work package in charge: WP 2 Scalability
• Actual delivery date for this deliverable: 1 August 2017
• Dissemination level: PU (for public use)
• Lead author: Deutsches Klimarechenzentrum GmbH (DKRZ): Philipp Neumann
• Other contributing authors: Deutsches Klimarechenzentrum GmbH (DKRZ): Joachim Biercamp, Irina Fast, European Centre for Medium-Range Weather Forecasts (ECMWF): Peter Bauer, Max-Planck-Institut für Meteorologie (MPI-M): Matthias Brück, Thorsten Mauritsen, Leonidas Linardakis, Deutscher Wetterdienst (DWD): Daniel Klocke