Computer Cluster for Numerical Modelling and Simulation

点击翻译按钮获取中文摘要

The working group Climate Geography focusses on numerical modeling of mountain glaciers on the regional and global scale, both for reconstructing past glacier dynamics, for producing projections and for conceptual studies of glacier-climate-interactions. The results are used for investigating the contribution of glaciers to seasonal water availability and for quantifying the contribution of glaciers to regional sea-level change. We also investigate societal consequences of these impacts of glacier change. Our work is based on the Open Global Glacier Model (OGGM, http://oggm.org), which is developed by the group with growing contributions from the international community. We further apply statistical and dynamical downscaling of the atmosphere to investigate interactions at the ice-air boundary, which are governed by the complex topography of mountain regions. Since the glaciers simulated by OGGM are not dynamically coupled, a global run of the model consists of a great number of independent processes, leading to very good scalability of the model. In principle, OGGM is therefore well suited to be deployed at regional/national high performance computation centers like HLRN. In practice, this is made impossible by the dependency of OGGM on a great number of external data sources, which are updated frequently and dynamically loaded from the internet. This is impossible at HLRN because of security considerations. Alternatively, data could be packaged and uploaded, which is impracticable because of the required amount of data (ca. 1 TB per global model run). Moreover, since OGGM is in very active development, it needs an up-to-date Python environment and frequent updates. This demand is incompatible with the compute centers’ (like HLRN) need for stability, which prevents users from getting direct control of the installed packages. Tests have shown that the dependence of OGGM on up-to-date software packages prevents its deployments at HLRN resources. Since model development is an important aspect of the ongoing work, local resources that can allocated and configured flexibly are needed. Finally, apart from the (well scalable) tasks of modeling a great number of individual glaciers, there are certain “global” tasks (mostly concerning model validation and optimization) that cannot be parallelized and thus need high maximum wall clock times. The compute cluster requested here is needed to secure the computational resources needed by the working group.