Characterizing and Improving Scientific Algorithms and I/O for Exascale: Nimble Dwarfs
Abstract
We employ elements of a "dwarf" method to characterize and influence the design of geofluid applications capable of running efficiently on diverse exascale systems. The basic requirements of our dwarf system is that it reflect a significant amount of the computational intensity, communication complexity and numerical accuracy of a full model while remaining small enough to facilitate rapid experimentation. This self-contained dwarf will be used as a framework to accelerate the testing and analysis of exascale-enabling technologies. We present a status of our dwarf development efforts for geofluid computation and I/O characterization. The former provide for examination of global geofluid modeling and computational issues using object-oriented components at low and high order truncation, and the latter enable characterization of distributed field I/O using a 'record-and-play-back' scheme.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMNG41B0950F
- Keywords:
-
- 0545 Modeling;
- COMPUTATIONAL GEOPHYSICSDE: 1942 Machine learning;
- INFORMATICSDE: 4430 Complex systems;
- NONLINEAR GEOPHYSICSDE: 4490 Turbulence;
- NONLINEAR GEOPHYSICS