Skip to Main Content U.S. Department of Energy
PNNL: High Performance Computing

High Performance Computing

In the last two decades, computing has been established as one of three methodologies of scientific research: theory, experimentation, and computational validation. However, exploiting hardware performance in science applications grew to be increasingly difficult. Additionally, approaches to grand challenge problems have historically focused on first-principle simulations, which start with a mathematical model, often expressed in terms of system of equations.

But the nation is facing problems that cannot be solved in the reasonable future using first-principle approaches. In biology, experimental data must infer biological networks and pathways. In national security, social networks must be derived from intelligence data. In energy supply, sensor networks must build real-time models of the national power grid network. These applications require a different system architecture to enable data-intensive computing.

Data-intensive computing starts from analysis and interpretation of massive amounts of data—data that are needed to build models and constrain the space of feasible models that make simulations computationally tractable. These data sets are much too large for effective storage, manipulation, archiving, navigation, visualization, and understanding.

PNNL is successful in high performance computing by merging multiple areas of science and technology.

  • Hardware: Processor speed, memory b/w, interconnect b/w, secondary storage, and reliability.
  • Algorithms: Scalable, resource-efficient, computational complexity, data decomposition, space/time locality, and load-balancing.
  • Software: Programming model, portability, numerical libraries, communication libraries, compilers, and debuggers.

Our researchers are enabling high performance computing for solving scientific problems by developing and implementing high-level programming abstractions. For example, our Global Arrays Toolkit provides a high-level, easy-to-use programming model with abstractions suitable for the science domains it targets. We are also developing problem-solving environments to increase ease of use and availability of high performance computing to non-specialists.

Site Links

Our Research

Past Research