Characterizing Large-Scale Computational Physics

Large-scale computational physics calculations typically share some of a number of basic characteristics: \begin{itemize} \item Brute-force approaches: Atomistic molecular dynamics, particle-in-cell plasma physics, particle-mesh cosmological simulations, DNS of turbulence, lattice QCD, Monte Carlo, {\ldots}. \item Wide range of relevant scales: Angstroms to millimeters in molecular dynamics, ion/electron cyclotron period to seconds or minutes in plasmas, galaxy to observable universe in cosmology, high Reynolds number turbulence, {\ldots}. \item Obvious need for yet larger scale: higher resolution, larger simulation domain, more particles, {\ldots}. \item Code is named, parallel, community, long-lived (but evolving). \end{itemize} This talk views the computational physics landscape from the perspective a physicist who has worked at three DOE large-scale computing centers: the Argonne Leadership Computing Facility, the (former) Advanced Computing Laboratory, and NERSC. The ``usual suspects'' at the large-scale end of computational physics are remarkably persistent, even in the face of an ever-increasing definition of large-scale.