Research activities in applied parallel
computing area at EMSL focus on: interprocessor communications, high-performance
input/output, programming models for hierarchical memory systems. One of
the primary goals for this work is to aid development of real scientific
application codes. A number of portable libraries and tools have been developed:
-
Global
Arrays (GA) is a portable shared-memory programming toolkit widely
available on current MPPs and networks workstations.
-
The GA
Visualizer helps programmers design efficient task scheduling by animating
access patterns to sections of 2-dimensional arrays.
-
Disk
Resident Arrays (DRA) extend the GA Non-Uniform Memory Access (NUMA)
programming model to disk.
-
Memory
Allocator (MA) is a local memory manager/allocator with several useful
features not available in Fortran or C languages.
-
ChemIO
is a package consisting of three independent parallel I/O libraries for
high-performance computers. It was designed for computational chemistry;
however, the supported abstractions and features are general enough to
be of interest to other applications.
-
TCGMSG
is an efficient but limited in functionality (comparing to MPI) message-passing
library available on many current (and legacy) systems.
-
TCGMSG-MPI
is a portability layer between TCGMSG and MPI. It is recommended as a transition
library from TCGMSG to MPI for existing TCGMSG codes.
-
Programmable
Array Compiler (PAC) provides a language compiler
that cross-compiles CM-Fortran to a parallel code using Fortran 77 with
message passing.
|
One-sided communication on the
IBM SP
A joint project with IBM involved
development of the LAPI
library. More information available in the IPPS'98
paper.
Slides on using LAPI presented at the IBM Performance Optimization Workshop
in March, 1999 in San Diego are here.
Information on using LAPI is also available in an IBM "redbook".
Remote visualization of large arrays
stored on parallel filesystems
A prototype GUI can be accessed from
the PNNL network here.
Distant
I/O
DIO model provides one-sided access
to remote secondary storage in a parallel program. It allows independent
read/write access to files at remote processor locations. The current implementation
uses Active Messages and asynchronous I/O. More Information on DIO available
in an HPDC-7
paper.
ARMCI
portable remote memory copy library
ARMCI provides vector and strided
interfaces to optimize performance of remote memory copy operations for
non-contiguous data.
|
quantum chemistry
(example codes include NWChem,
GAMESS-UK,
Columbus,
Molpro,
Molcas)
molecular dynamics
(e.g., NWChem package)
computer graphics
atmospheric
chemistry
computational
fluid dynamics
financial security
forecasting (at Bear Stearns)
A joint IBM/PNNL patent no 6,070,189
"Signaling
communication events in a computer network" on technology developed
for LAPI communication library on the IBM SP, May, 2000.
R&D100 Award for MS3
package that includes NWChem, Ecce and Parsoft, July 1999.
Supercomputing'98 Best Paper award
Some of the papers are accessible
from
here.
The HPCC
publication list contains many references to papers written based on
work performed at PNNL with this software.
Contact Info: Jarek Nieplocha |
Maintained by J.
Nieplocha <j _ nieplocha@pnl.gov>/ 06.30.98
|