Workshop: Exploiting Different Levels of Parallelism for Exascale Computing

Important dates:

tba

Abstract

Today, many scientific computing applications feature different levels of parallelism. In some cases, this effect is induced by the increasing complexity of the underlying model. Here, one example are multi-physics applications if realized in a partitioned manner, e.g., with separate solvers for the involved physical models, which ca lead to inter- and inner-solver parallelism. In other cases, there are several levels of parallelism that are inherent to solver methods. E.g., combination techniques for high-dimensional systems of equations combine solutions on several grids to an overall so-called sparse grid solution in a smart way. Monte-Carlo methods sample a given space with values at random positions. This sampling is obviously embarassingly parallel while the values at the sample points can be solutions of large systems of equations that are computed using a parallel solver, too. In addition, in order to increase parallelism, parallelization in time in combination with parallelization in space has become an active field of research within the last years. In other cases, levels of parallelism are added deliberately to solver methods. Multigrid methods are for example combined with domain decompositions, or the Jacobi-Davidson method is modified in order to add some multi-scale parallelism. In most of theses cases, an open question is how to exploit this in the actual implementation on peta- and exascale architectures. Choices to be made here mostly concern the assignment of parallel work portions over all levels to hardware components which has severe influence in particular on the communication costs and, thus, on the overall scalability. The workshop is organized by the coordinators of the projects EXAHD and ExaFSA in the priority program 1648, Software for Exascale Computing, of the German Research Foundation. Thus, contributions from other projects of this priority program are particularly welcome.

We are looking forward to seeing you in Munich!

Miriam Mehl, Dirk Pflüger

Workshop: Exploiting Different Levels of Parallelism for Exascale Computing

as part of the ACM International Conference on Supercomputing - ICS 2014

Date

June 10, 2014

Venue

Bavarian Academy of Sciences, Munich

Contact

Miriam Mehl, miriam.mehl@ipvs.uni-stuttgart.de
Dirk Pflüger, dirk.pflueger@ipvs.uni-stuttgart.de