Thursday, March 1, 2012

Supercomputer Powers NASA’s Climate Research

Analyzing the world’s climate and creating global climate models demands a supercomputing
capability that continually pushes the leading edge. Since 2000, CSC has helped the NASA Center for Climate Simulation (NCCS) operate, maintain and improve its supercomputing systems.
 
Currently, the NCCS Discover computing cluster ranks in the top 100 of the world’s supercomputers
and is a leader among those systems focused on climate and weather research.
 
The center, based at NASA’s Goddard Space Flight Center, integrates supercomputing,
visualization and data interaction technologies to support research for more than 500
scientists at NASA centers, and researchers at laboratories and universities around the world.
 
“The computer is the climate scientist’s tool — the better the tool, the better the scientific
results, and the greater the understanding of what’s happening in the complete Earth
system,” says Phil Webster, head of Goddard’s Computational and Information Sciences
and Technology Office. “A key challenge for us is to build better machines because what
we need doesn’t exist.”
 
In the past five years, CSC has helped increase Discover’s performance 130-fold. Today,
the NCCS computing cluster uses more than 35,000 processing cores to crunch more
than 400 trillion floating-point operations per second. By comparison, it would take every
person on Earth adding pairs of seven-digit numbers at the rate of one per second for
more than 17 hours to do what Discover can do in one second.
 
Managing big data
 
Another challenge for the center, says Webster, is data management, or more accurately,
Big Data management. Scientists using the center integrate millions of observations collected daily, reanalyze past observations and perform climate model simulations,
each of which can produce massive amounts of data. CSC also helps administer
Discover’s archive system, which stores about 28 petabytes1 of data, with a total
capacity of 37 petabytes.
 
“The Big Data problem is like finding a needle in a needle stack,” says Scott Wallace,
CSC NCCS Support program manager. “Finding your needle in a pile of 28 trillion
needles is not significantly harder than finding it in a pile of one trillion needles because
they’re both effectively impossible, unless you build in a way to keep track of where each
needle is located.”
 
As the center generates and manages increasing quantities of data, it has turned to
visualization technologies to help scientists see their research. A recent addition to the
center is its Visualization Wall, driven by16 Linux-based servers. These servers split
images across the 17-by-6-foot wall, creating one huge, high-resolution medium on
which scientists can display still images, video and animated content from data
generated on Discover.
 
“The wall gives scientists an important new tool because it lets them see their research in
incredible detail,” says Fred Reitz, CSC NCCS Support operations and deputy program
manager.
 
A keener focus
 
Even as the center improves its capabilities, researchers continue to ask for more. For
example, several groups of scientists have more than doubled their workload requests
because of upcoming deadlines on key projects such as the Fifth Assessment Report for
the Intergovernmental Panel on Climate Change (IPCC), the leading international body
for climate change assessment.
 
Today, Discover can compute in one day three simulated days in the life of the Earth at
one of the highest resolutions ever attained—about 3.5-km global resolution, or about 3.6
billion grid cells. The center’s current “stretch” goal is to generate in one day a computation
that covers 365 days at 1-km global resolution.
 
“Just in terms of electricity, that one computation would require 16 megawatts2 of power
the way things are done today,” says Wallace. “This isn’t within reach now, but that’s our
distant grail. We’re forever looking for better resolution and faster times.”
 
Recently, in fact, the center reached a new benchmark when Discover ran the highest
resolution atmospheric simulation of its kind, modeling two years of the Earth’s climate
at 10 km globally. To achieve advances like these, NASA also taps CSC’s High
Performance Computing Center of Excellence for assistance.
 
Established in 1999, the CSC center has more than 160 specialists operating systems
that collectively provide capacity for more than 110 petabytes of data and have a capability
of almost two petaflops3 of computation.
 
“Climate research continues to stretch computing capabilities,” says Donna Klecka, director
of CSC’s High Performance Computing Center of Excellence. “Through our center, we
can further support NASA’s center, bringing our deep computing expertise to innovate
and create Big Data solutions that its climate scientists need.”

No comments:

Post a Comment