Science-of-Scale
Projects
To promote the productivity of cutting-edge
science, NERSC adapts its systems to provide the resources needed
by data-intensive applications and provides “Red Carpet”
support to large-scale strategic projects so that they can make
rapid progress. These projects typically need support such as
very large computing allocations, very large scratch disk, terabytes
of usable memory and support for 64-bit computing, large-scale
visualization, consulting support to make effective use of these
resources, good bandwidth between the resources, and large archival
storage.
The projects described
below show how the unique resources of the IBM SP combined
with excellent consulting support have produced scientific
breakthroughs. They also demonstrate the size of the resources
needed for rapidly growing, data-intensive projects. Three
of these projects could have used all of NERSC’s resources
in FY 2002.
DOE Climate
Change Prediction Program
Scientists in the DOE Climate Change Prediction Program recently
completed a 1,000-year run of a powerful new climate system
model on a supercomputer at NERSC. The millennium-long simulation
of the new Community Climate System Model (CCSM2) ran for
more than 200 uninterrupted days on the IBM SP supercomputer
at NERSC. The lengthy run served as a kind of “shakedown
cruise” for the new version of the climate model and
demonstrated that its variability is stable, even when run
for century-after-century simulations. The 1,000-year CCSM2
run had a total drift of just one-half of one degree Celsius,
compared to older versions with two to three times as much
variance.
|
|
|
|
Figure 1
Two hundred years of modeling El Niño events and
surface temperatures on the Community Climate System Model
(CCSM2) closely correlate with 50 years of actual climate
data. |
|
A 1,000-year simulation demonstrates the ability of CCSM2
to produce a long-term, stable representation of the earth’s
climate (Figure 1). Few if any climate models in the world
can make this claim, since all previous simulations contained
drifts too large to allow complete, uncorrected simulations
to 1,000 years. In addition, the simulation provides scientists
with a database to analyze the variability of weather and
climate on time scales ranging from interannual to interdecadal
to intercentennial. Few datasets exist which are as comprehensive
as the one produced during this simulation.
CCSM2 tightly couples four complex models, including atmosphere
and land modeling codes developed at the National Center for
Atmospheric Research (NCAR) and ocean and sea ice models developed
at Los Alamos National Laboratory. Computationally, the full
CCSM2 code consists of five binaries which are organized to
execute concurrently within a single job. The models exchange
data at various frequencies appropriate to the physical, large-scale
processes being simulated. CCSM2 requires 4.5 wall-clock hours
on 144 1.5-Gflops CPUs of the NERSC IBM SP to complete one
simulated year. NERSC gave CCSM2 special queue priority to
complete this project in a timely fashion. Preliminary results
of 800 model years were presented to 250 participants at the
Seventh Annual CCSM Workshop held in Breckenridge, Colorado,
on June 25–27, 2002.
INVESTIGATORS
W. Washington, P. Gent, J. Hack, J. Kiehl, G. Meehl, and P.
Rasch, National Center for Atmospheric Research; B. Semtner,
Naval Postgraduate School; J. Weatherly, U.S. Army Cold Regions
Research and Engineering Lab Laboratory.
PUBLICATIONS
J. T. Kiehl and P. Gent, “The Control Climate Simulation
from the Community Climate System Model (CCSM2)” (in
preparation).
URL
http://www.ccsm.ucar.edu/index.html
|