SC²S Colloquium - Nov 8, 2018
|Date:||Nov 8, 2018|
|Time:||15:00 - 16:00|
Raffael Düll: Evalulation and Adaptation of the Flexible Node-Level Library AutoPas in Molecular Dynamics Simulations
This is a Bachelor's Thesis submission talk advised by Fabio Gratl
The aim of this Bachelors thesis is to analyse the performance and the scalability of the developing AutoPas library for molecular dynamics. A testing program which supports pairwise intermolecular interactions for atoms and rigid molecules has therefore been developed . The evaluation of the library is performed on thin nodes of the SuperMUC phase 1 and on CoolMUC-3 at the LRZ Linux-cluster. The main aspects of this thesis where OpenMP parallelisation on a single node and vectorisation by SIMD instructions. It has been shown that the efficiency of the force calculations with the AutoPas library improves with adequate vector instructions. At node-level, AutoPas scales well if few cores are used, but the execution time does not decrease as rapidly as expected for a higher number of threads. The overall performance remains below the performance of the highly scalable program ls1-mardyn.
Keywords: AutoPas, Molecular Dynamics, Node-Level Performance
Hendrik Möller: Dimension-wise Spatial-adaptive Refinement with the Sparse Grid Combination Technique
This is a Bachelor's Thesis submission talk advised by Michael Obersteiner
Spatial adaptive refinement with the sparse grid combination technique is an effective way to adapt the combination technique to non-smooth functions. One problem of the standard combination technique is that all grids are always regular, making local refinement impossible. This paper tackles this problem by presenting the use of a dimension-wise spatial-adaptive refinement strategy. It also shows how error estimation was tweaked in order to increase effectiveness and multiple variants tested with the presented algorithm and compared to the standard combination technique.
Keywords: Sparse Grid Combination Technique, Spatial adaptivity, curse of dimensionality