Dresden 2017 – wissenschaftliches Programm
CPP 11.1: Vortrag
Montag, 20. März 2017, 15:00–15:15, ZEU 255
Scalable and fast concurrent multiscale molecular simulation with predictive parallelization schemes — •Horacio Vargas Guzman1, Christoph Junghans2, Kurt Kremer1, and Torsten Stuehn1 — 1Max Planck Institute for Polymer Research, Ackermannweg 10, 55128 Mainz, Germany — 2Computer, Computational, and Statistical Sciences Division, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
Concurrent multiscale simulation enables the study of molecular systems with different resolutions in specific subdomains of a simulation box. Modeling soft-matter and biological systems in the context of multiscale simulations are challenging research avenues which drive the permanent development of new simulation methods and algorithms. In computational terms, those methods require parallelization schemes that make productive use of computational resources for each simulation and from its genesis. Here, we introduce the dual resolution domain decomposition algorithm that is a combination of a resolution sensitive spatial domain decomposition with an initial sliding subdomain-walls procedure. The algorithm modeling is presented for dual resolution systems in terms of scaling properties as a function of the size of the low-resolution region and the high to low resolutions ratio. The algorithm competences are validated within adaptive resolution simulations, by comparing its scalability and speedup to a spatial domain decomposition. Two representative adaptive resolution simulations have been employed in this work, namely, a biomolecule solvated in water and water in an ideal gas reservoir.