Transcript Document

The Materials Computation Center, University of Illinois
Duane Johnson and Richard Martin (PIs), NSF DMR-03-25939 • www.mcc.uiuc.edu
Quantum Monte Carlo Applications for
Petascale Computers
Objectives: To develop and test efficient quantum
Monte Carlo (QMC) simulation codes for breakthrough
calculations of electronic systems on petaflop-scale
computers.
Approach: Quantum Monte Carlo has multiple
possibilities for parallelism, so that use of terascale
facilities has not required special attention to attain
good scalability. But scaling up the parallelism another
two orders of magnitude will require exploiting more
avenues for parallelism; for example, utilizing the
multicore/shared memory nature of the nodes to
parallelize a single iteration of random walk. In
addition, careful attention to fault tolerance and load
balance is required.
Significant Results: Development and release of
highly scalable and efficient QMC suite (QMCPACK,
PIMC++). Performance enhancement using
OpenMP/MPI hybrid programming on multi-core
systems.
Broader Impact: The developed tools will be
available to the scientific community through the open
source project QMCPACK.
Principal investigators: D.M. Ceperley and J. Kim
(a) Overall parallel efficiency of Diffusion Monte
Carlo (DMC) simulations; (b) efficiency of OpenMP
implementation of Variational MC (VMC) and DMC. The
performance analysis was performed on NCSA Intel 64
Cluster Abe.