• HomePage
• Research Team
  • Phil Colella (lead PI)
  • Jim Demmel (co-PI)
  • John Mellor-Crummey (co-PI)
  • Vivek Sarkar (co-PI)
• Workshops
  • Combustion, Sep 23-25, 2013
  • Climate, Oct 14-16, 2013
  • Plasma Physics, Nov 15-17, 2013
  • Biology, Mar 11-13, 2014
  • Astrophysics and Cosmology, May 7-9, 2014
• WikiSandbox

PmWiki

• Documentation Index

edit SideBar

• View
• Edit
• History
• Print

HomePage

Institute for High-Performance Computational Science with Structured Meshes and Particles (HPCS-SMP)

HPCS-SMP is a conceptualization effort for an HPC simulation software Institute across multiple science grand challenge areas funded by NSF’s Software Infrastructure for Sustained Innovation (SI2) Program. Through a series of workshops with domain science simulation experts, we are identifying opportunities to leverage software infrastructure, expertise and collaboration between software developers and application scientists, to evolve new HPC grand challenge simulation capabilities in their fields. HPCS-SMP is looking at opportunities across six different general scientific application areas:

• Astrophysics and cosmology, including star formation, large-scale evolution structure of the universe, and energetic events (supernovae and x-ray bursts).
• Spatial modeling in biology, both at the cellular scale (models for reactions at low concentrations) and at the macroscale (continuum mechanics, chemical transport).
• Combustion for engineering systems, including multiphase effects (fuel sprays) or radiative heat transfer (low-emissions industrial burners), and multi-physics design problems.
• Atmospheric and subsurface geophysical fluid dynamics, including the atmosphere and oceans, moisture dynamics and chemical species; radiation; and the biosphere; hydrology and carbon sequestration, dynamics of the earth's crust, mantle and core.
• Plasma physics and kinetics, addressing large-scale dynamics of charged / ionized systems, coupled to electromagnetic fields, and kinetic systems with long mean free paths.

In general, these science domains are described by PDEs - fluid dynamics, electromagnetism, transport phenomena, continuum mechanics - coupled at multiple scales, often with particle methods, and share many of the same patterns for simulation, algorithm and software developments that can be reused across structured grid and particle applications. However, grand challenge-type simulation capabilities will require a unique software infrastructure to adapt to the disruptive changes of emerging and next-generation HPC architectures. As we observe the rise of HPC computing systems employing graphic processing units (GPUs) or Intel many integrated core (MIC) processors as accelerators, applications will need to embrace heterogeneous processor cores, multi-threading, fine-grain data-parallelism, and partitioned memory spaces. Even more dramatic challenges are anticipated for exascale architectures that are expected at the end of the decade: more parallelism within individual nodes; severe resource limitations (memory, bandwidth, etc); heterogeneous cores and deep hierarchical memory systems; and roughly 1000x more frequent faults than today’s systems. Therefore, applications will need to be redesigned to exploit an explosion in parallelism, manage scarce resources, handle failures, and dynamically adapt to changing needs and contexts. This entails fundamental and unavoidable changes to scientific simulation and approaches for developing applications.

The goal of HPCS-SMP is to identify the components - mathematical algorithms combined with software frameworks, libraries and tools - and the opportunities for an ongoing collaboration between application scientists and software developers to realize new simulation capabilities that are well-suited for emerging and future HPC systems.

Additional reading:

• The Future of Computing Performance: Game Over or Next Level?, Samuel H. Fuller and Lynette I. Millett, Editors, Committee on Sustaining Growth in Computing Performance, National Academy of Sciences, 2011. (PDF version free for downloading.)
• The Heterogeneous Programming Jungle, Michael Wolfe (Compiler Engineer, The Portland Group, Inc), HPC Wire, March 19, 2012.
• The Potential Impact of High-End Capability Computing on Four Illustrative Fields of Science and Engineering, Committee on the Potential Impact of High-End Computing on Illustrative Fields of Science and Engineering, National Academy of Sciences, 2008. (PDF version free for downloading.)
• The Opportunities and Challenges of Exascale Computing, DOE ASCAC Subcommittee Report, November 2010.
• Multiphysics Simulations: Challenges and Opportunities, International Journal of High Performance Computing Applications, February 2013, 27: 4-83, doi:10.1177/1094342012468181.