Hardware Reference
In-Depth Information
Since the 1992 moratorium on nuclear testing, computer simulation has
had an increasingly important role in this stockpile stewardship. The goal of
these simulation efforts is to provide scientists and engineers with the technical
capabilities to maintain the nuclear weapon stockpile without the key tool that
had been applied for 50 years: physical testing of nuclear devices.
The Advanced Simulation and Computing program (ASC, formerly known
as the Accelerated Strategic Computing Initiative, ASCI) was established in
1995 to develop this simulation capability. ASC is creating these capabilities
using advanced codes and high performance computing that incorporates more
complete scientific models based on experimental results, past tests, and the-
ory. To meet these needs, ASC is solving progressively more dicult problems.
Applications must achieve higher resolution, higher fidelity, three-dimensional
physics, and full-system modeling to reduce reliance on empirical evidence.
This level of simulation requires computing resources at ever-increasing scales
and levels of performance. To accomplish this, ASC collaborates with industry
to accelerate development of more powerful computing systems and invests in
creating the necessary software environment.
LC is home to a first-class computational infrastructure that supports the
computing requirements of LLNL's research scientists and collaborating sci-
entists from around the world. A particular focus of LC is to develop solutions
(in collaboration with the \Tri-Lab" partners at Los Alamos and Sandia Na-
tional Laboratories) that will create a functional problem-solving environment
for high performance computers under the ASC program [1]. Another goal is
to provide leveraged, cost-effective high performance computing to multiple
programs and independent researchers under the Multiprogrammatic and In-
stitutional Computing (M&IC) program [9].
LC has been involved with bringing high performance parallel file systems
into production since the inception of the ASCI Initiative. LC's rst experi-
ences with IBM's General Parallel File System [6] were with the initial ASCI
systems. They had increasingly large instances of GPFS through the pro-
gression of the ASC machines: Blue-Pacific (2.1 TFLOPS), then White (7.3
TFLOPS), and finally Purple (76 TFLOPS). The GPFS file system on ASC
Purple was the largest installed parallel file system at the time with 2 PB of
deployed storage and a bandwidth capability of 100 GB/s (which could be
delivered to a single application program).
In order to achieve the I/O requirements and the related components of
the Livermore usage model for the ASC and M&IC program, LC collaborated
with other sister DOE National Laboratories at Los Alamos and Sandia. The
Tri-Labs began work in 1999 with industrial partners Cluster File Systems,
Inc. (founded by Peter Braam), Hewlett{Packard, and Intel to develop the
Lustre
R
file system with the 1.0 release in 2003.
As a part of the procurement that brought ASCI Purple to LC, LLNL
also began work with the Blue Gene series of computers from IBM. The Blue
Gene series of systems is distinguished from previous ASC systems not only
