Automated
Instrumentation and Monitoring System (AIMS)
AIMS
is a software toolkit for capturing, displaying, and analyzing
parallel program behaviors and performance. AIMS has three major
components: a source code instrumentor that automatically inserts
active event recorders into program source code before compilation;
a run-time performance monitoring libarary that collects performance
data; and a visualization tool set that reconstructs program execution
based on the data collected. AIMS developers now use the toolkit
as the basis for creating new instrumenting and monitoring techniques,
and is being incorporated into various hardware testbeds to evaluate
their impact on user productivity.
Big
Dataset Visualization
NAS
researchers are investigating two approaches to the interactive visualization
of very large datasets. In the first, memory hierarchies, virtual-memory-like
paging techniques allow standard visualization programs to load only the currently
needed data. To hide the delays caused by the retrieval of data from local
or remote disks, researchers are developing algorithms for multitasking. The
second approach, multi-resolution representation, transforms data into wavelet
representations, which naturally allow multi-resolution browsing as well as
high memory and bandwidth savings.
Charon
Parallelization Toolkit.
Charon
is a library of C and Fortran functions that partially automate the transformation
of legacy software for structured-grid simulations into parallel, portable,
scalable codes based on the Message Passing Interface. It makes domain decomposition
less cumbersome by creating and manipulating distributed arrays, with the
effect of relaxing the restriction of data arrays to individual processors.
When Charon was used to parallelize the Scalar-pentadiagonal section of the
NAS Parallel Benchmarks, the resulting program had only 20 percent more code
and improved its performance from 61 MFLOPS on one processor to 2.7 GFLOPS
using 81 processors.
Commodity
Computing
Because
of their low cost, networked clusters of PCs offer greater cost/performance
ratios than traditional supercomputers. NAS researchers are building and testing
prototype commodity clusters, developing performance models for applications
in areas such as computational fluid dynamics, and designing system architectures
that take advantage of commodity technology. In 1999 they completed the design
of a scalable network topology, the P-mesh, based on a grid of switches. In
tests, P-mesh has proved to be just as fast as, but less failure-prone and
expensive than, other network topologies such as trees and toruses. With continued
research, commodity clusters may become a major supplier of computational
power for the Information Power Grid.
Computational
Stellar Dynamics
The
behavior of galaxies containing hundreds of billions of stars can be modeled
in large time-varying N-body simulations. NAS researchers are developing novel
approaches to visualizing the results of such simulations. By combining a
Smoothed Particle Hydrodynamics (SPH) method, time-varying volume visualization,
an interactive transfer-function editor, and a time-varying isosurface routine,
it is possible to visualize and investigate processes, such as galaxy formation,
which normally occur over large expanses of space and time. In a simulation
of the passage of a black hole through a galactic core, for example, the researchers
discovered unexpected oscillations, density enhancements and depletions, and
gravitational wakes in the stellar field.
DAO
Origin 2000 Migration Project
The
Data Assimilation Office at NASA Goddard Space Flight Center owns a cluster
of SGI Origin 2000 computers at the NAS Facility, used mainly to process data
collected by Earth Observing System missions such as POLARIS (Photochemistry
of Ozone Loss in the Arctic Region in Summer) and STRAT (Stratospheric Tracers
of Atmospheric Transport). New parallel versions of GEOS-DAS, the custom data
analysis software used by DAO, require more memory and more floating point
operations per second than the older serial versions. Part of the consulting
support provided to DAO by the NAS Systems Division involves improving the
speed, efficiency, and stability of the new GEOS-DAS code.
Dynamic
Load Balancing Techniques for Distributed Computing Applications
Large-scale
parallel applications such as flow solvers can benefit from load balancing
tools developed at NAS to improve the efficiency of computations. For instance,
a portable system called PLUM (Parallel Load Balancing for Adaptive Unstructured
Meshes) redistributes the computational mesh to underutilized processors,
whenever the computations associated with an individual mesh partition or
processor rise significantly above average. Efficient mapping and execution
of parallel codes is key to the success of parallel computing applications
in distributed computing environments.
Earth
Observing System (EOS) Data Analysis and Modeling
NASA
is deploying a fleet of satellites carrying innovative sensor systems to gather
atmospheric, oceanic, and earth science data that, over the long term, will
help scientists to develop a model of the Earth as a unified system. For example,
the low-orbit ICESat mission, to be launched in 2001, carries a laser altimeter
designed to measure changes in ice-sheet topography at the Earth’s poles.
As part of the EOS Atmospheric Physics and Land Surface program at Goddard
Space Flight Center, supercomputers at the NAS Facility are used to create
climate models from data gathered by these remote sensing satellites. NAS
scientists are also helping EOS mission scientists port a legacy serial climate
modeling code (the Goddard Earth Observing System Data Assimilation System,
or GEOS-DAS) from the Cray vector supercomputers to the SGI Origin 2000 shared-memory
computers. As the Information Power Grid matures, it will be used to transfer
EOS datasets transparently from the point of collection (Goddard) to the point
of computation (the NAS Facility) and back to the point of analysis (Goddard).
MPI-2
I/O
MPI-2
I/O (Message Passing Interface-2 Input/Output) is the first and only standard
interface supporting file access for parallel computing applications. As the
computing community adopts the standard, applications based on MPI will become
easier to port between architectures. NAS researchers helped to develop the
standard in collaboration with Argonne National Laboratory and the MPI Forum.
Multi-Source
Data Analysis
The
VISOR system
(Visual Integration of Simulated and Observed Results) is a client-server
testbed application for the visualization and analysis of data from multiple
sources such as wind tunnel measurements and computational fluid dynamics
simulations. It maps data from disparate sources to a common 2-D or 3-D
framework and helps extract relevant data from large datasets. A template
language update under development will allow client-side customization
of the data structure transmitted by the server. VISOR researchers are
working with colleagues from the Ames DARWIN project.
NAS
Parallel Benchmarks (NPB) and Programming Baseline for NPB (PBN)
The
NAS Parallel Benchmarks
are standardized performance tests for parallel computers. The benchmarks
include eight computational fluid dynamics problems that help scientists and
managers compare the performance of their computers with those at other sites.
Version 2.3 of NPB was released in 1996 and has remained unchanged, providing
a fixed reference point. Now NAS is offering a comparable set of problems,
the Programming Baseline for NPB (PBN),
designed under the directives-based OpenMP, High Performance Fortran, and
Java programming paradigms. A beta version of PBN was released in 1999 and
PBN 3.0 is expected to be released in 2000.
Portable
Parallel/Distributed Debugger (p2d2)
p2d2
is a graphical tool for debugging up to 128 processes running in parallel
or on heterogeneous collections of high speed computers. Some of the goals
of the p2d2 project are: to provide a user interface that is consistent across
all platforms and is scalable to thousands of processes; to accommodate standard
debugging operations such as breakpoints, data watchpoints, single-stepping,
and data displays; and to permit debugging irrespective of the communication
library used, such as MPI or PVM. P2d2 has been used to debug heterogeneous
computations running under the Globus computational grid environment. It is
expected that a beta version of p2d2 will be distributed domestically in early
2000.
Quantum
Molecular Topology
By
computing and visualizing mathematical properties of the fields of electrical
charge around atoms and molecules, researchers at NAS and in the Department
of Chemistry at Middle Tennessee State University have devised a new way of
representing atoms, molecules, and their interactions. In this new understanding,
atoms are conceived as the spaces bounded by certain "zero-flux surfaces"
where the electron charge density is neither increasing nor decreasing. The
work synthesizes recent advances in quantum chemistry with vector field topology
and computer graphics feature detection techniques originally developed for
aerodynamics data. It helps to explain why molecules take their distinctive
shapes, how they orient themselves for chemical reactions, and even where
new atoms may bind if a molecule is bent out of shape. The techniques have
already led to a new understanding of the mechanism and geometry of hydrogen
bonding, particularly in biomolecules such as deoxyribonucleic acid (DNA).
Surface
Flow Visualization
A
thin layer of oil applied to an aircraft model before wind tunnel tests will
make complex surface flow patterns visible. NAS researchers have developed
software that automatically generates the same patterns on lifting bodies
in simulated flow fields. The software depends on a texture synthesis technique
known as Line Integral Convolution (LIC). It can be used to highlight features
such as separation and attachment lines, which are important warning signs
of vortices and other aerodynamic phenomena that can reduce lift.
UFAT
A
part of the Visualization Techniques Library, UFAT
computes particle traces and other features in datasets representing time-dependent
physical flow. These features are then displayed using a graphical utility
such as the Flow Analysis Software Toolkit (FAST). Researchers can use UFAT
to pick out path lines, streak lines, stream surfaces and ribbons, and color-contoured
grid surfaces. For example, NASA aerospace engineers working on SOFIA, an
infrared astronomy instrument borne aloft by a Boeing 747, used streak lines
computed by UFAT to assess the aerodynamic impact of the large cavity housing
the telescope.
Virtual
Mechanosynthesis
Using
the Virtual
Mechanosynthesis (VMS) application, users can see, move, and even feel
simulated molecular structures in three dimensions. The software couples an
accurate molecular dynamics simulation code with an immersive graphical display
that has interactive capabilities and manual force feedback. Atoms appear
as softball-sized spheres and can be "grabbed" and moved about the
scene at will. The force feedback device lets the user feel the attractive
and repulsive forces as the atom pushes and pulls on its neighbors, and the
neighboring atoms respond to the same forces with physically realistic movements.
VMS may eventually help researchers explore, rehearse, and debug precise assembly
sequences for future nanoscale devices.
Visualization
Techniques Library
The
VisTech
Library is a collection of reusable software tools designed
to automatically detect and highlight different features of 2-D
or 3-D datasets, such as those produced in time-varying computational
fluid dynamics simulations. The library, which facilitates rapid
prototyping of visualization systems, includes routines for visualizing
grid surfaces, streamlines and streaklines, separation lines,
shock surfaces, vortex cores, isosurfaces, cutting planes, vectors,
cells, and other flow field features
|
|
