HPC System with High Memory and IO Requirements

• Term: Funded in 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 440395346

An increasing number of institutes of Kiel University have very high demands for compute and storage resources to drive their scientific simulations and data analysis, both for research and teaching. Large parts of this workload cannot be handled by traditional parallel HPC resources, as these typically only have limited memory per CPU core. For instance the field of life sciences, e.g. bioinformatics and ocean research, shows increasing demand for simulations and analysis of huge metagenomics data. These studies are often only moderately parallel and are characterized by their huge memory and storage requirements.Algorithms with very high memory consumption are increasingly emerging from other disciplines as well. In theoretical chemistry quantum mechanical descriptions are used, which require the diagonalization of huge matrices. These algorithms are difficult to parallelize and work with huge data sets, which are used repeatedly - resulting in massive speedups if the matrices can be kept in memory completely (research group B. Hartke). In theoretical condensed matter physics the situation is comparable. Algorithms for the simulation of correlated systems, e.g. nonequilibrium Green functions, or path integral Monte-Carlo methods for quantum plasmas or Fermi fluids, are memory intensive and profit immensely from large main memory (research group M. Bonitz). The self-consistent simulation of electron-light interaction in nano-optics is currently not possible at all on the available systems with low to medium memory (research group N. Talebi).From the 13 research agendas submitted 11 propose very focused applications. Two agendas cover the field of method development in numerics. The group of M. Braack develops finite element methods for the efficient solution of partial differential equations and optimal control problems. The group of A. Srivastav examines efficient applications of streaming algorithms and big data problems in exponentially large search spaces, based on current problems of groups in the "Kiel life science" and "Kiel marine science" context.The HPC resources currently available at Kiel University are very heterogeneous - financed piecewise from different institutes and projects over a long period and operated either from the computing center or locally by the institutes. This results in problems with the efficiency of operation and with overall utilization of the available hardware. The main goal of this proposal is the replacement of the largely obsolete hardware and the consolidation of the resources into a single, large scale cluster, which is to be used by all research groups involved. At the same time the proposed system is going to be an addition to the local tier 3 HPC resources, providing the research groups at Kiel University with the essential prerequisites of scientific simulation and for teaching and training in collaboration with the computing centre.

• DFG Programme: Major Research Instrumentation
• Major Instrumentation: HPC System mit hohen Speicher- und IO Anforderungen
• Instrumentation Group: 7040 Vektorrechner
• Applicant Institution: Christian-Albrechts-Universität zu Kiel
• Leader: Professor Dr. Anand Srivastav