Results

HPC Weather benchmarking suite. HPCW isolates key elements in the workflow of weather and climate prediction systems to improve performance and to allow a detailed performance comparison for different hardware platforms thus fostering co-design with vendors and technology providers

Containerization of the code EC-EARTH for running climate simulation (comes from AUTOSUBMIT WF manager and EC-EARTH)

Application area: Climate simulations in pre-exascale computers (continue -> Codes involved: EC-Earth consortium Efforts undertaken to achieve exascale readiness for each particular code/showcase: Containerisation is one potential approach to port EC-Earth to new systems, which includes pre-exascale systems. Given that access to tier 0 systems usually comes with severe time constraints and strong focus on actual production runs, container based porting can help to better utilise the assigned resources. Design and testing of the containerisation approach takes portability of the computational performance into account.

Tools to reduce the precision of the variable model automatically from double to single

Tools to analise the computational performance of an application automaticaly

porting / integration on MN5 and LUMI

Use of PSyclone in the NEMO build system. During the first 18 months of the project the goal is to use PSyclone to port the NEMO code on GPU-based platforms with two configurations of interest for CMCC and namely a local configuration tailored on the mediterranean region at 1/24° horizontal resolution and GLOB16 which is a global configuration including sea-ice at 1/16°. PSyclone is also used by the official NEMO release and it is part of the NEMO build system. In the last part of the project the goal is to provide feedbacks to the PSyclone developers providing optimised OpenACC code that could be used to improve the PSyclone toolchain. Finally, PSyclone will be improved adding code check functionality to automatically verify whether the NEMO code (and its future updates) is compliant with the RPE tool and hence ready for mixed precision.

This is the new dynamical core of ECMWF that will be used for operational weather predictions and climate simulations. The model has been developed outside of ESiWACE but ESiWACE is supporting the porting of the dycore to GPU hardware and heteoregeneous hardware via the GT4Py DSL

Evaluation of the readiness of published climate simulation output for automated anaysis. High potential because automated analysability of climate data has not been evaluate before and because specifications to enhance/enable machine actionability can be deduced from this work

The Field Compression Library can be used by scientific groups to explore data compression for weather and climate science

As part of the first ESiWACE3 hackathon, at CSC om 18.10.-20.10.2023, the application of data compression online as a simple toy model (Lorenz96) is running was explored. This preliminary experiment showed that (1) data compression can be beneficial in reducing I/O load of the model, which no longer has to save uncompressed data to disk, (2) extraneous lossy data compression alters the behaviour of a model ensemble s.t. the result can no longer be used, (3) lossy data compression within a safe zone produces a model ensemble with equivalent behaviour even when compression is applied to the internal model state at every timestep. Therefore, if lossy compression is performed within this safe zone, models can be resumed from compressed states without altering their statistic ensemble behaviour. This result could be used to drastically lower the storage and data transfer costs for model restart states.

The Online Laboratory for Climate Science and Meteorology (https://lab.climet.eu) provides a JupyterLab-like environment that has common climate and meteorology packages and libraries (e.g. eccodes) preinstalled. Unlike a typical JupyterLab instance which requires a server to host all running Python sessions, the online lab is serverless (just a static website) and runs entirely inside the user’s webbrowser using WebAssembly. It builds on the existing Pyodide and the JupyterLite projects, which provide Python/JupyterLab in the webbrowser using WebAssembly. Our contribution are (1) that several weather and climate packages can now run in this environment and (2) several transparent-to-the-user patches to these and other packages to ensure that the same code runs in the online lab just as it would in a local installation. Overall, the online lab provides researchers an installation-free quick-to-launch Jupyter environment in their web browser that can be used for running small experiments, sharing notebooks, and hosting interactive documentation. For example, the Online Compression Laboratory (https://compression.lab.climet.eu) hosts the compression laboratory notebooks (R31) using the online laboratory so that anyone can get started trying out compression on weather and climate data without any setup in ~1min. The initial prototype (-2023/11) was primarily focused on the technical infrastructure of the online lab, i.e. getting critical Python libraries for weather and climate science to work in the new environment (see above) and implementing data streaming to support working with large datasets in the memory-constrained webbrowser environment. The initial releases v0.1.0 and v0.2.0 provided support for additional packages, improved user convenience, and added the needed functionality to host the online compression laboratory.

The (field) compression laboratory builds on ECMWF’s field compression library. It provides several Jupyter notebooks that explore data compression for weather and climate data. The notebooks can be run locally or in the Online Laboratory (R30) (https://compression.lab.climet.eu).

The Kernel Tuner Ecosystem is a set of tools and libraries to improve the perfrormance of GPU software.