Exascale Scientific Applications: Scalability and Performance Portability PDF

Preface

Scientific computing has become one of the fundamental pillars of science, combining theory and experiment. Computing is providing capabilities allowing theoretical concepts to be cast in computational modeling and simulation methods for the interpretation, prediction, and design of experiments or for providing unique and detailed understanding of physical systems that are impossible or prohibitively dificult, expensive, or dangerous to study experimentally. Computing also plays an increasingly important role in the analysis of large-scale observational and experimental data with the objective of validating or improving the theoretical models of the underlying physical phenomena, as well as informing and guiding new experiments. The scientific enterprise is depending on computing to address many of the fundamental intellectual challenges for understanding the natural world including the evolution of life, the properties and reactivity of materials that make up our environment, and the formation and expansion of the universe. Computing has an increasingly transformational role in practically every aspect of society as well, including economic competitiveness, advanced manufacturing, health care, environmental sustainability, natural disaster recovery, social media and entertainment, national security, and energy security.

The enormous advances in the integration of computing into virtually everything we do is in part the result of the rapid technological developments of the last decades. The largest computers available have become faster by almost three orders of magnitude roughly every decade. Current leadership computing facilities provide systems capable of providing tens of petaops and exaopscapable systems are expected in the 2021–2023 timeframe. The computer architectures that have made these increases in processing power possible have gone through a number of significant conceptual changes, from fast scalar processors in the 1970s, vector processors in the 1980s, parallel systems in the 1990s and 2000s, to the current transition from massively parallel homogeneous computer systems to the highly complex systems with extensive hierarchies in processors and accelerators, volatile and nonvolatile memory, and communication networks.

Authors: Tjerk P. Straatsma, Katerina B. Antypas, Timothy J. Williams