Topic 16: Extreme-Scale Computing

Description

Following seven orders of magnitude improvement in performance on scientific and engineering applications over the past 24 years – which has raised expectations for predictive simulation as a tool of discovery, design, and decision support – the road to extreme performance is encountering different challenges and becoming much steeper. Traditionally, our scientific computing code base is focused on optimizing floating point operations and improving the execution rate of those that remain, but this will not suffice in the future. On one hand, diverging exponentials in hardware subsystem performance require more attention to parallel programming. High concurrency and power-efficient design of the individual cores bring opposite pressures: greater data locality and greater freedom to redistribute data and computation.
For reasons of energy efficiency and system acquisition cost, we must now focus on squeezing out synchronizations, total memory footprint, and limiting memory transfers. On the other hand, scientific applications are become more complex, for example, in order to address multiscale/multiphysics, reinforcing the requirements for numerical accuracy, synchronization and resiliency. Rethinking programming models, algorithmic implementations, and even mathematical models preferred as starting points will therefore be major milestones on the way to extreme scale.

We welcome papers charting the path of extreme simulation and data analytics in science and engineering onto emerging architectures, at all levels of the modeling chain.

Focus

  • Co-design of applications, algorithms, and architectures
  • Synchronization-reducing and communication-reducing algorithms
  • Programming models for distributed-shared hybrid architectures
  • Execution models and execution monitoring for performance optimization on distributed-shared hybrid architectures
  • Machine learning techniques applied to performance optimization
  • Case studies of scientific and engineering simulations and data analytics focused on exascale architectural constraints

Topic Committee

Global chair
David Keyes, King Abdullah University of Science and Technology, Saudi Arabia

Local chair
Marie-Christine Sawley, Intel Exascale Lab Paris, France

Further members
Thomas Schulthess, ETH Zurich, Switzerland
John Shalf, Lawrence Berkeley National Laboratory, USA

News

Workshop proceedings published

The workshop proceedings have been...  more

Keynote slides online

The slides of all three keynote...  more

Conference app available

The conference program of Euro-Par 2013...  more

Sponsors:

-->