Teaching

Numerical Mathematics (Engineering Science, Master)

Parallel Computing for Multiscale and Multiphysics Problems (Engineering Science, Master)

Programmieren und Datenanalyse mit Python (Teil 2) (ISA)

Special Applications of HPC in Defence Technology (Engineering Science, Master)

Introduction to Computer Science (Engineering Science, Bachelor)

HPC Techniques and Software Development (Engineering Science, Master)

Programmieren und Datenanalyse mit Python (Teil 1) (ISA)

Advanced Numerical Mathematics (Engineering Science, Master)

Hardware Architecture of HPC Systems (Engineering Science, Master)

Practical Training (Engineering Science, Bachelor)

Numerical Mathematics (Engineering Science, Master)

Parallel Computing for Multiscale and Multiphysics Problems (Engineering Science, Master)

Special Applications of HPC in Defence Technology (Engineering Science, Master)

Hardware Architecture of HPC Systems (Engineering Science, Master)

Programming – Part 1: Introduction to Computer Science (Engineering Science, Bachelor)

If you are interested in pursuing a student project, feel free to directly our team!
Project topics are defined on an individual basis, depending on your interests and our current research foci.

Current projects:

  • Comparison of Performance Prediction Methods
    The performance of numerical simulations depends typically on many parameters – time step, number of grid points, vectorization properties, tolerances for iterative schemes within the simulations, and so forth.
    To schedule simulations efficiently on HPC clusters and, thus, to save energy costs, accurate predictions for the run time of a simulation under a particular parametrization would be desirable.
    This project shall compare various methods for data analysis on given run time data. The corresponding evaluation shall be established via Python and shall be based on a first evaluation script that has been used in a prior performance prediction study [NEU19].
    Prerequisites: Some experiences in programming, preferably in Python.
    References:
    [NEU19] P. Neumann. Sparse Grid Regression for Performance Prediction Using High-Dimensional Run Time Data. Euro-Par 2019: Parallel Processing Workshops, 2019 (accepted)
  • Towards Fault Tolerance in Molecular-Continuum Simulations
    Molecular-Continuum Simulations couple molecular dynamics (MD) and CFD methods to explore fluid flow close to the molecular scale where a pure MD simulation would still be too computationally expensive.
    Executing these simulations on large-scale supercomputers however is challenging. One issue is given by some singular, accidentally broken compute cores out of millions of them that are assembled in the supercomputer.
    This project shall extend the macro-micro-coupling tool (MaMiCo) towards fault tolerance. MaMiCo employs multi-instance sampling, that is many randomized MD instances are coupled to one CFD simulation. This project shall extend the communication pattern and in particular the MPI communicators used in MaMiCo to react in a fault-tolerant way if one compute core crashes during execution. This particularly includes the incorporation of the User Level Failure Mitigation, an extension of MPI to modify MPI communicators.
    At the end of the project, partitions with a crashed compute core should be automatically excluded from MaMiCo and removed from MD sampling.
    Prerequisites: Experiences in MPI programming, C/C++.
    References:
    [NEU17] P. Neumann, X. Bian. MaMiCo: Transient Multi-Instance Molecular-Continuum Flow Simulation on Supercomputers. Computer Physics Communications 220, pp. 390-402, 2017
  • Feature Tracking in Molecular-Continuum Simulations
    Molecular-Continuum Simulations couple molecular dynamics (MD) and CFD methods to explore fluid flow close to the molecular scale where a pure MD simulation would still be inhibitively computationally expensive.
    Recently, noise filters have been incorporated in MaMiCo to allow for more efficient flow signal extraction from the highly fluctuating MD data [JAR19].
    This project shall use the interfaces from the noise filter approach to incorporate more data analysis methods into MaMiCo, such as vorticity detection, max/min flow velocity detection, etc.
    Prerequisites: Experiences in C/C++. MPI knowledge is a helpful, but not necessary prerequisite.
    References:
    [JAR19] P. Jarmatz, P. Neumann. MaMiCo: Parallel Noise Reduction for Multi-Instance Molecular-Continuum Flow Simulation. ICCS 2019 proceedings, LNCS 11539, pp. 451-464, 2019

Further directions for projects comprise (but are not limited to):

  • Coupling molecular dynamics packages or CFD solvers with our coupling tool MaMiCo
  • Implementation of and performance analysis for particle simulations
  • Implementation of straight-line Lattice Boltzmann solvers on different hardware architectures

HSU

Letzte Änderung: 14. Januar 2021