High-Performance Computing

The problem sizes and data amounts used for scientific studies of the SDL Climate Science require the use of the High-Performance Computing environment at the JSC.

The SDL Climate Science is also involved in the exploration and development of new software paradigms. A key feature for performance is a proper memory layout and management especially on heterogeneous architectures, which often leads to different data structures and code paths. For this reason, Domain Specific Languages (DSL) have been successfully introduced in several scientific disciplines. Theoretically, they offer a proper way of abstraction from the underlying hardware, while allowing for improved readability within the domain they are used in. The concept of an embedded DSL was adapted from previous work in ParFlow and implemented in EULAG.

The SDL Climate Science is also involved in the investigation of the modular supercomputing architecture (MSA). The changing challenges in the field of high-performance computing (HPC) require innovative strategies to enable scientifically demanding as well as high-performance computing in the exascale range. On the hardware side, one approach is to develop supercomputers according to the modular design principle. Such HPC systems take into account a modular component approach connected by a powerful network and a cross-system software and programming environment. The modular supercomputing architecture is since end of 2020 in production (Suarez et al. 2019). The JUWELS system has a general purpose CPU cluster and an upscaling GPU booster. Scientific applications can choose which module they need and use one or more of it. We are using the MSA approach for coupled simulations of the ICON model. The atmospheric part is running on the booster and the ocean and I/O part is running on the cluster.