"We expect this system to be a
big hit, but we must prove it to ourselves and our users."
SCD director Al Kellie
On 12 July 2004, SCD took
delivery of "lightning," a large-scale, high-performance
Linux cluster manufactured by IBM. (See photos, below.)
The acquisition is part of SCD's new five-year Strategic Plan to evaluate new technologies and find ways for SCD to deliver
more cost-effective tools for advancing the frontiers of science at
The 1.1-teraflop system is considerably faster on a per-processor
basis than bluesky, NCAR's IBM Cluster 1600. In benchmark tests, the
Community Atmosphere Model (CAM) tests ran 30% faster, while the Parallel
Ocean Program (POP) ran 40% faster. One reason for this is that lightning's
Opteron processors, built by Advanced MicroDevices (AMD), have much
better memory bandwidth than bluesky's Power4 processors.
"I believe the time is right for NCAR to evaluate the applicability
of commodity-based Linux clusters for their readiness in an enterprise
computing environment like SCD's," says SCD director Al Kellie.
"The press is full of success stories, but very few of them stress
the operating environment, equipment performance, and reliability as
much as we do. We expect this system to be a big hit, but we must prove
it to ourselves and our users."
Many of NCAR's university partners who want to use models such as
the Community Climate System Model (CCSM) or the Weather Research and
Forecast (WRF) model have Linux-based systems. Lightning will give
them, for the first time, the chance to build, test, and evaluate these
codes in a full-scale Linux computing environment similar to their
Dick Valent, assistant manager of SCD's User Support Section, concurs.
"This system will give us valuable experience with the new type
of Linux platforms that are going to be showing up at more and more
Delivery and configuration
Lightning was assembled and tested by IBM in its facility in Guadalajara,
Mexico. It was shipped to Boulder by truck and arrived at NCAR's Mesa
Laboratory the morning of Monday, 12 July. SCD and IBM staff worked
with Eagle Global Logistics, the moving company, to unload the five
frames (total weight: 9,870 pounds) and wheel them through the halls
to the Computing Room.
Once the frames were in place, IBM engineers began hooking up nodes,
attaching wiring, and connecting power. SCD and IBM are currently working
on system configuration and software installation, monitoring the equipment
as it runs to check vital signs such as temperature and air flow.
In the days to come, Jim Tuccillo is slated to run benchmark tests
on behalf of IBM, ensuring that lightning meets performance specifications.
Rory Kelly of SCD's Computational Science Section will then rerun the
benchmarks on behalf of NCAR. Once the benchmarks have been validated
by both IBM and NCAR, Steve Gombosi of SCD's User Support Section will
begin acceptance testing, running codes 24 hours a day for a week to
subject the machine to heavy loads and check for correct answers. If
the machine is accepted, friendly-user testing will follow.
Lightning is expected to be made available to the larger user community
sometime this Fall.
A divisional effort
Project manager Georges Fuentes, head of SCD's Supercomputer Systems
Group, notes that the installation and support of the new Linux cluster
is a divisional effort involving the hard work and contributions of
many SCD staff. Core project team members and the sections they represent
- Director's Office: Tom Bettge, Tom Engel, Rich Loft
- Computational Science: Rory Kelly, Henry Tufo
- High-Performance Systems: George Fuentes, Mark Genty, Gene Harano,
- Network Engineering and Technologies: Marla Meehl, David Mitchell
- Operations and Infrastructure Support: Aaron Anderson, Gary New
- User Support: Ginger Caldwell, Bo Connell, Dick Valent,