Max Planck Institute Magdeburg puts new high-performance Linux cluster into operation
The Linux cluster mechthild with 132 nodes, 2192 cores, 27 terabyte main memory and 63 trillion arithmetic operations per second supports the researchers at the Max Planck Institute for Dynamics of Complex Technical Systems Magdeburg in their simulations and computations since August 2018.
In the recent years, the need for more central computational resources has grown steadily at the Max Planck Institute Magdeburg. The new supercomputer mechthild replaces the computing system otto which was installed in 2010. The various research groups at the MPI can use it for their large and, especially distributed, computations and development of scientific software – autonomously, unlimited in time and with sixfold computational performance, while at the same time only consuming double the energy.
The largest part of mechthild consists of 112 nodes for general computations. The ten big memory nodes are used for memory demanding algorithms. According to the requirements of their applications the users can load various preinstalled software packages.
The Molecular Simulations and Design group at MPI, headed by Dr. Matthias Stein, intensively uses the cluster. They are investigating the molecular and structural basis of complex mechanisms and processes in chemistry and biology. The systems studied range from chemical reactions in solution, enzymatic reactions, self-assembling systems, proteins in membranes to computer-aided drug design. The scalar relativistic density functional theory (DFT) calculation of the electronic structure in mixed-valence transition metal complexes is a very expensive operation. Determining the spectroscopic observables (IR, EPR) and the additional consideration of thermally accessible conformers requires high-performance hardware.
A team of researchers of the MPI group Physical and Chemical Foundations Process Engineering wants to understand the molecular-level origins which determine retention in hydrophilic interaction liquid chromatography (HILIC). This causes expensive simulations. Despite of growing commercial use of HILIC the retention mechanism behind is not yet clear. Simulations using at least 120 processors help them to achieve new results in a reasonable period of time.
For the researchers in the scientific computing team of the Computational Methods in Systems and Control Theory group the unique feature of the new cluster is the possibility to monitor the power consumption of the entire machine on a per node basis with high accuracy and time resolution. Having investigated power saving algorithms on single compute servers, together with their colleagues in the Max Planck Partner group at UdelaR in Montevideo (Uruguay), mechthild enables them to extend their investigation towards distributed computations.
The Max Planck Institute and the Max Planck Society invested around 900.00 EUR into the installation of mechthild.