![description or caption](images/71.jpg) |
Transition
to parallel computing computers
from left to right: Cray x-MP,
Cray 2, Cray c-90, Paragon, CM-5,
Cray T3-E, Nirvana Blue, ASCI
Red, IBM SP |
Supercomputers are plenty fast, but
a number of them working in parallel
are orders of magnitude faster. Researchers
at several national laboratories and
universities supported by the Office
of Science produced the software,
scalable operating systems, and other
technologies needed for massively
parallel supercomputing (involving
1,000 or more processors) and demonstrated
its value in fields ranging from seismic
imaging to materials modeling. In
one example, Oak Ridge National Laboratory
and collaborators developed Parallel
Virtual Machine (PVM) software, which
allows heterogeneous collections of
computerseven personal computers
and workstationsto be linked
together regardless of location, and
treated as one parallel computer.
This software, first released publicly
in 1991, led to the development of
cluster computing, in which many inexpensive
machines are connected to create a
powerful system. Argonne National
Laboratory developed software that
provides a portable environment for
building, running, and examining the
performance of programs in a wide
variety of parallel computing environments.
The combined work on massively parallel
systems has won three Gordon Bell
awards and six R&D 100 awards from
R&D Magazine recognizing
significant new technologies, and
received numerous patents.
Scientific Impact:
Massively parallel computing can solve
problems up to 100 times faster than
serial supercomputers. PVM has tens
of thousands of users in scientific
and other fields and has become the
de facto global standard for distributed
computing; thousands of programmers
use the Argonne software, which increases
productivity by enabling applications
development on diverse architectures
without program performance losses
or time-consuming code changes.
Social Impact: The
growing popularity of parallel computing
for industrial and medical applications
is exemplified by wide use of PVM,
which helps many large companies design
new products cost effectively. PVM
also is used as an educational tool,
enabling universities without access
to parallel computers to teach parallel
programming courses.
Reference: van der
Weide, E.; Deconinck, H.; Issman,
E.; Degrez, G., "A parallel, implicit,
multi-dimensional upwind, residual
distribution method for the Navier-Stokes
equations on unstructured grids,"
Computational Mechanics,
Mar 25, 1999, ISSN 0178-7675.
Using MPI, Portable Parallel Programming
with Message Passing Interface,
Gropp,W; Lusk, E.; Skjellum,A, MIT
Press, Cambridge, MA (1994).
ScaLAPACK Users' Guide, L.S.
Blackford, J. Choi, A Cleary, E. D'Azevedo,
J. Demmel, I. Dhillon, J. Dongarra,
S. Hammarling, G. Henry, A. Petitet,
K. Stanley, D. Walker, R.C. Whaley,
SIAM, Philadelphia (1997).
URL:
http://www.sc.doe.gov/production/octr/
Technical Contact:
Daniel A Hitchcock, Mathematical,
Information, & Computational Sciences
Division, 301-903-6767
Press Contact: Jeff
Sherwood, DOE Office of Public Affairs,
202-586-5806
SC-Funding Office:
Office of Advanced Scientific Computing
Research |