Linux cluster for geoscientific capacity simulations

在过去的几十年里，计算机模拟对于理解地球内部复杂过程的重要性急剧增加。这需要永久提供具有高计算能力的模拟基础设施。多年来，LMU的地球物理学主席一直专注于地球科学建模，如今涉及高性能计算和数据爆炸的相互作用，即来自测量或模型计算的数据的巨大增长。研究的重点是模型与观测的定量结合。为此，应用了数值上极其复杂的数据同化方法。由于许多地球物理问题本质上是逆问题，计算要求高的迭代优化方法用于从给定的观测集确定复杂的模型状态。如今，并行高性能计算机使求解具有超过10x10个自由度的离散系统成为可能。具体来说，这意味着我们可以在三个数量级上模拟地球物理过程的尺度相互作用。Linux计算集群可以很好地满足计算基础设施的最终需求。本文所应用的系统作为一个高容量的模拟吞吐量系统是必不可少的。它的目的是覆盖主要容量计算需求，并进行系统的地球物理参数研究，这可能需要几个月的计算才能完成。因此，该系统是LMU地球物理讲座模拟基础设施的核心组成部分。为了确保这一基础设施的持续长期安全性，有必要对2012年获得的当前Linux集群进行替换。我们预计，在这个快速增长的高吞吐量（容量计算）的中型计算部分（中尺度计算）中，未来将会有更多的计算需求。

The importance of computer simulations for understanding complex processes within the Earth has grown dramatically over the last few decades. This requires the permanent provision of simulation infrastructures with high compute capacity. For many years the chair of Geophysics at LMU has been focused on geoscientific modeling, which nowadays involves the interaction of high-performance computing and data explosion, i.e. the immense growth of data coming either from measurements or model calculations. The research is focused on the quantitative merging of models and observations. For this purpose, numerically extremely complex data assimilation methods are applied. Since many geophysical problems are inverse problems by nature, computationally demanding iterative optimization methods are used to determine complex model states from given sets of observations. Parallel high-performance computers make it possible today to solve discrete systems with more than 10exp10 degrees of freedom. Specifically this means that we can model the scale interaction of geophysical processes over three orders of magnitude. The resulting requirements for the computational infrastructures are particularly well met by Linux computing clusters. The system that is applied for here is essential as a simulation throughput system of high capacity. It is intended to cover the primary capacity computing needs and to perform systematic geophysical parameter studies, which may take many months of computation to complete. The system is thus a central component of the simulation infrastructure of the geophysics chair of LMU. To ensure the continuing long-term security of this infrastructure, the intended replacement of the current Linux cluster, acquired in 2012, is essential. We anticipate that there will be even more computing demand in the future within this rapidly growing mid range computing segment (mesoscale computing) of high throughput (capacity computing).