SCCS Colloquium - Oct 21, 2020

From Sccswiki
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
Date: Oct 21, 2020
Room: Online (BBB)
Time: 15:00 - 16:00

Wolf Thieme: Parallelization of Existing Tuning Strategies in AutoPas using MPI

Bachelor's thesis submission talk. Wolf is advised by Fabio Gratl.

There exist many different strategies to solve Molecular Dynamics (MD) simulations. Choosing which to use for a given simulation requires unrealistic foresight, making it desirable to choose them automatically, which is called auto-tuning. AutoPas is a C++ MD-simulation library designed to do exactly that with custom simulations provided by the user. Because every AutoPas process in a parallel simulation previously had to find an optimal strategy in isolation, this thesis implements a parallelization for the tuning process using the MPI-specification. As the solution has to be applicable for all processes, the simulation-domain is assumed to be homogeneous. Testing the implementation in several situations shows that significant improvements in performance are possible, but also that they are highly situational. Therefore, future optimizations will be necessary to fully exploit the potential of parallelizing the tuning.

Sebastian Schäfer: Efficiency Analysis of Seismic High Performance Simulations

Bachelor's thesis submission talk. Sebastian is advised by Sebastian Wolf and Dr. Anne Reinarz.

High parallel efficiency is an important goal in high performance computing (HPC) environments. The seismic simulation software SeisSol and the PDE solver engine ExaHyPE both try to achieve this goal. Measuring and comparing the parallel efficiency is important to give a better insight on how far the goal is reached. In this thesis, two seimsic scenarios were simulated and the parallel performance was measured. The talk will present SeisSol and ExaHyPE and will present the results of the performance measured.

Keywords: High Performance Computing, Seismic Simulations, Parallelisation