Argonne National Laboratory

The New Paradigm of
Petascale Computing

(Page 3 of 3)

Similarly, Argonne is teaming with François Gygi at the University of California at Davis and his research team to examine the fundamental physical nature of nanomaterials that gives them unique properties. One of the biggest challenges in nanoscience, according to Gygi, is learning how to control nanostructure assemblies.  The scientists want to understand and control the interacting of nanoscale building blocks used to design new materials.  The computational challenge, then, is to predict interface properties at the nanoscale.

Gygi, in collaboration with Giulia Galli (also at UC Davis) wants to use BlueGene/P to study silicon dots embedded in silicon nitride and carbon nanotubes in solution.  The investigation will focus on whether it is possible to build a silicon laser out of dots assembled on a matrix.  The application runs on Qbox, a program built on an Message Passing Interface (MPI) framework and calculating first principles molecular dynamics.  It is one of a handful of programs that have already demonstrated the ability to take full advantage of Leadership Computing systems.

In 2007, Qbox set the world record for floating-point performance by achieving a sustained performance of 207 teraflops on the LLNL BlueGene/L 65,536 node system. That corresponds to 56.5 percent of the theoretical peak using all 128,000 CPUs.  Gygi’s team simulated the electronic structure of molybdenum, a high-Z or heavy metal, that is important to the National Nuclear Security Administration’s (NNSA) Stockpile Stewardship Program.  In this case, the study was the first step toward simulating the effects of aging on nuclear materials.

Future Programming

Capitalizing on their strengths in scientific computing, teams of Argonne researchers are already working on next-generation programming. Current linear solvers, a key component of scientific software, are reaching their limit in scalability, says Stevens.  As part of DOE’s SciDAC-funded project “Towards Optimal Petascale Simulations (TOPS).” Argonne researchers, led by Lois McInnes, are working to relieve the linear solver bottleneck. Similarly, through SciDAC’s Center for Technology for Advanced Scientific Component Software, Argonne scientists are working to expand the software toolbox for scientific discovery to include programs that act as discrete components with unique capabilities, but that also work together seamlessly.  The idea is to build a component ecosystem, a community of interacting units that work together and provide feedback to each other in a self-regulating way.

Stevens says that most of these projects are in their beginning phases and there are many places where a DOE CSGF fellow could plug into the organization.

“It’s an exciting time to be starting a career in scientific computing,” Stevens says.  “Depending on the career interests of the person, there are opportunities to become a co-developer, a discipline scientist, a numerical analyst, a visualization expert.  It’s all open right now.  The structure of leadership computing means that you can have close collaborations between teams over time.  That’s really the vision and the promise of this approach.”

« Previous       1   |   2   |   3   |   References   |   Print       Next »