The vision

EXCLAIM

The OpenETH project EXCLAIM aims at developing EXtreme scale Computing and data platform for cLoud-resolving weAther and clImate Modeling, an external pageICON-model based infrastructure that is capable of running kilometer-scale climate simulations at both regional and global scales. This platform will become a foundation for weather and climate modeling for the decades to come.

The challenge

Numerical simulations of the atmosphere and climate system rely on complex software – hardware infrastructure with codes containing millions of lines that represent the underlying physical equations. Reaching our objective requires not only a two to three-orders of magnitude increase in computational throughput, but also efficient workstreams and data-flows that deal with the resulting data avalanche. In the development of this infrastructure, special attention will be given to its usability and transferability.

Creating this infrastructure requires a highly collaborative effort that brings together computer scientists with data and domain experts, using innovative coding concepts and exploiting the next generation of supercomputer architectures and the power of novel hardware systems including accelerators such as Graphics Processing Units (GPU). EXCLAIM will also very closely work together with its national and international partners.

Impact

The urgency for high-resolution climate models is driven by the need of combatting climate change and developing effective measures to adapt to the unavoidable part of climate change ahead of us. An improved representation of extremes such as heatwaves, intense rainfalls, and droughts, in terms of their intensity, duration and recurrence will support also much more accurate assessments of their impacts on people, critical infrastructure and ecosystems. High spatial resolution is also required to better simulate weather and climate change impacts on sensitive infrastructures and ecosystems, especially in areas with complex terrain, such as the Swiss Alps, or for urban areas. The EXCLAIM implementation of the extreme scale computing and data platform will support the next generation of external pageSwiss Climate Change Scenarios, currently scheduled for about 2027.

Implementation

The core approach taken by EXCLAIM to reach the performance target is a fundamental redesign and refactoring of the model codes from the current primarily Fortran-based paradigm to one that enables performance portable, architecture agnostic user codes. This involves splitting of the currently monolithic codes into a descriptive user code (based on Python) and into a backend that translates to a standard imperative language (e.g., C/C++) for specific architectures (here especially heterogeneous GPU/CPU based architectures). In addition, increases in the performance of a number of numerical schemes, new architectural concepts such as coarse-grained reconfigurable arrays, and the expected increase in memory and hardware performance will help to reach the global kilometer-scale objective within the framework of EXCLAIM.

For the refactoring, EXCLAIM follows a non-disruptive approach that enables scientific productivity during the entire project. This is ensured by a stepwise approach, where individual components of the model system are replaced by the Python-based framework (GT4Py) one by one, and then tested thoroughly before advancing to the next step. Continuous integration/continuous deployment (CI/CD) methods will be used to ensure the full functioning of the built systems along the way. A well designed
verification and testing system ensures scientific validity of the results.

For more information see our high-level implementation plan.

Planned evolution of ICON from Fortran/OpenACC to Python/GT4Py. The steps are: (1) explore the use of Python/GT4Py with one or a few parameterizations; (2) replace the dynamical core of ICON with a Python/GT4Py-based implementation; (3) to make available the whole (or most) of ICON in Python.
Planned evolution of ICON from Fortran/OpenACC (blue) to Python/GT4Py (green). The steps are: (1) explore the use of Python/GT4Py with one or a few parameterizations; (2) replace the dynamical core of ICON with a Python/GT4Py-based implementation; (3) to make available the whole (or most) of ICON in Python.
JavaScript has been disabled in your browser