home 2007 talks 2006 talks 2005 talks 2004 talks abstracts
Vitali Morozov, Computer Simulation and Numerical Methods of Detail Atomic
Physics Calculations for Various Applications
A key requirement in the simulation of dynamics, thermal, optical,
and transport properties of laboratory and natural plasma sources is
self-consistency since each involved physical process influences the properties
of the others. We have been developing comprehensive and integrated models and
environments for plasma simulation for various applications. These simulations
incorporate several models for the calculation of basic one-electron atomic
properties, plasma electron concentration and partition function, thermodynamic
properties and equation-of-state, opacities and resolution of radiation
transfer. In this talk, I will present the full path of what we are trying to do
and the challenges we come across in our computer simulation of plasma
applications. These challenges range from modeling one-electron wave functions,
energy levels and transition probabilities by Hartree-Fock/Slater methods,
through calculation plasma balance and populations of levels within the
framework of Collisional-Radiation Equilibrium model to generating opacities and
resolution Radiation Transport Equation (RTE). Several algorithms of solving RTE
in parallel computers are presented and discussed.
Stephen Thomas, IMAGe, NCAR: Aqua-planet experiments with a spectral element GCM
We describe an atmospheric general circulation model based on
spectral elements. The dynamical core was initially coupled to a simplified
physics package including a moist convective parameterization due to Emanuel. An
idealized aqua-planet experiment proposed by Grabowski and Moncrieff was then
run on the IBM Blue Gene/L supercomputer at up to 10 km horizontal resolution.
Results obtained on a 32 rack system (65,536 processors, 183.5 TeraFlops peak)
achieve sustained performance levels of up to 11.3 TeraFlops More recently we
have coupled the spectral element dynamical core to the full NCAR CAM physics
package and have performed more extensive aqua-planet simulations. Transient
space-time analysis indicates the presence of quasi-periodic precipitation
structures similar to the MJO.
Jose Luis Morales, ITAM: An algorithm for the approximate and fast solution of
mixed symmetric LCPs
In this talk we discuss an algorithm for the solution of large scale LCPs. The
proposed algorithm combines iterations of projected Gauss-Seidel method with
subspace minimization steps.
Fred Stevens, BIO.
Extreme Protein Homology Search
Most pairs of proteins that share a common evolutionary past --
homologs -- do not exhibit statistically significant amino acid sequence
identity. Therefore, the ability to generate hypotheses of function by the
exclusive use of methods dependent on statistical significance is limited and
restricts assignment of function by combined computational and experimental
strategies.
An opportunity exists to develop and implement homology search algorithms that
are based on a "forensic" analysis of genomic data, methods we have used to
identify unsuspected binding sites in the cancer-associated BRCA1 and to suggest
that apparent convergent evolution of structure exhibited by several beta-domain
proteins may in fact be divergent evolution of function from a common ancestral
protein.
Pete Beckman,
MCS, ZeptoOS, BG/P, and Petascale Machines
In the beginning, supercomputers ran a simple operating system. In some cases,
customers wrote their own OS for the raw hardware. As MPPs and clusters emerged,
they began to run thousands of instances of an operating system. Today,
supercomputers have several types of operating systems on different functional
units: storage nodes, compute nodes, login nodes, service nodes, etc. Most of
the time they are based on Linux. As machines scale up, however, simple design
choices what worked well for servers or several thousand nodes suddenly become
bottlenecks. Can Linux be used for petascale BlueGene/P systems? Are preemptive
multitasking operating systems doomed to fail on petascale machines? We
investigate noise by
measuring existing operating systems and injecting artificially generated noise
into a massively parallel system to measure its influence on the performance of
collective operations. What are the fundamental constraints, and what does the
real data suggest? This talk will address these questions and present data
gathered from experiments on the Watson BG/L system. The ZeptoOS Linux kernel
and build system for BG/L I/O nodes will also be presented, as well as our plans
for running Linux on BlueGene/P compute nodes.
Todd Munson, MCS, The Perfect STRM
Globalized Newton methods minimize a merit function along a direction
to obtain the next iterate. When the algorithm converges to a local minimizer
that is not a global minimizer of the merit function, the convergence rate
deteriorates. We look at using second-order information for the merit function
to improve the convergence rate of the globalized algorithm to these local
minimizers. Some preliminary experience with a semismooth trust-region method (STRM)
for solving complementarity problems on multiplayer matrix games will be
presented.
Tim Tautges, MCS, Local Topological Modification of Hexahedral Meshes Using
Dual-Based Operations: Progress and Application
Local topological modification is widely used to improve mesh quality after
automatic generation of tetrahedral and quadrilateral meshes. These same
techniques are also used to support adaptive refinement of these meshes. In
contrast, few methods are known for locally modifying the topology of hexahedral
meshes. Most efforts to do this have been based on fixed transition templates or
global refinement. In contrast, a dual-based "pillowing" method has been used
which, while local, is still quite restricted in its application, and is
typically applied in a template-based fashion. In this presentation, I will
describe the generalization of a dual-based approach to the local topological
modification of hex meshes and its application to clean up hexahedral meshes.
A set of three operations for locally modifying hex mesh topology has been shown
to reproduce the so-called "flipping" operations described by Bern et. al as
well as other commonly-used refinement templates. I will describe the
implementation of these operators and their application to real meshes.
Challenging aspects of this work have included visualization of a hex mesh and
its dual (especially for poor-quality meshes); the incremental modification of
both the primal (i.e. the mesh) and the dual simultaneously; and the interactive
steering of these operations with the goal of improving hex meshes which would
otherwise have unacceptable quality. These aspects will be discussed in the
context of improving hex meshes generated by curve contraction-based whisker
weaving. Application of these techniques for improving other hexahedral mesh
types, for example those resulting from tetrahedral subdivision, will also be
discussed.
Paul Hovland, MCS:
Hierarchical Coloring
Discretizing PDEs on a regular grid induces a hierarchy of two types
of sparsity patterns on a Jacobian matrix, one due to the stencil being used and
one due to the dependence among individual degrees of freedom within that
stencil. Goldfarb and Toint demonstrated how to exploit the sparsity structure
induced by the stencil, but exploiting the unstructured sparsity within the
stencil must rely upon more general techniques. We present a two stage coloring
strategy: Goldfarb-Toint coloring followed by optimal intra-stencil coloring. We
demonstrate that optimal coloring at both stages is suboptimal in general.
Nonetheless, savings of 50% or more over Goldfarb-Toint alone are possible.