Arctic Region Supercomputing Center

Delivering Teraflops of HPC Power for Arctic Research with Sun Fire X2200 M2 Server and Sun Fire X4600 Server

The Arctic Region Supercomputing Center at the University of Alaska Fairbanks operates under the U.S. Department of Defense High Performance Computing Modernization Program (HPCMP). Located on the campus of the Arctic university of the United States, ARSC is a research leader in supporting scientific analysis and discovery of Polar Regions with state-of-the-art technology and high-performance computational resources.

Customer Challenges

Provide 24x7 access to computational resources for worldwide community of Arctic researchers and HPCMP users
Provide a cost-competitive solution for supercomputing in a complex, heterogeneous environment
Provide high performance computing resources to scientists and students
Incubate future generations of computational scientists focusing on Polar Region research

Solution

Sun delivered a factory-built high-performance computing (HPC) cluster solution built around Sun Fire X2200 M2 and Sun Fire X4600 servers racked, stacked, cabled and tested, providing a completely integrated hardware solution with very competitive performance.

Business Results

The remote, partial assembly saved time so that on-site deployment could focus on system configuration, software level testing and operating systems
Enabled ARSC to more accurately predict weather changes to make air travel safer
Cemented ARSC’s position at the forefront of the nation’s university-based supercomputing centers, with focused computational support for polar research
Created the potential for successful, international study of Arctic Systems Modeling
Enabled, planned and optimized computing performance at teraflop speed
Placed the first Sun supercomputer in the HPCMP program

Products and Services

Story Details

Based at the University of Alaska Fairbanks (UAF), the Arctic Region Supercomputing Center (ARSC) has distinguished itself as a pioneer in high performance computing and data storage capabilities for polar research. ARSC has pledged to its scientists more than 1 million CPU-hours per year on “Midnight,” a Sun Opteron cluster, that support earth systems research. Midnight is now a key resource for modeling the atmosphere to provide greater detail to the air traffic controllers and weather forecasters who guide pilots flying the skies. Significant computational resources using Midnight have also been committed to an intensive Arctic System Modeling effort (ASM). To develop ASM, UAF’s International Arctic Research Center and ARSC are working together to couple land, ice, ocean and atmospheric models and applying the coupled system to studying changes in the Arctic.

Each compute Opteron core in the cluster as 4-gigabytes of system memory. Midnight’s smaller x2200 nodes have four processor cores on two sockets, while the larger x4600 nodes feature 16 cores on eight sockets. Across numerous HPC applications and benchmarks, the x2200 nodes were found to be more efficient per core providing a significant speed advantage over many other machines for large memory jobs. This speed helps to bring weather-related research to fruition at a faster pace.

" With the Sun HPC solution, ARSC is able to provide the computational basis for research and instruction that requires high resolution modeling of the atmosphere including regional weather, prediction of the location and duration of effects of smoke from wildfires, and forecasting the dispersion of ash from volcanic eruptions. "

— Dr. Gregory Newby, Chief Scientist, Arctic Region Supercomputing Center

The Midnight supercomputer is comprised of:

2 Sun Fire X4600 nodes which act as login nodes with 32 gigabytes of shared memory per node; 4 dual core AMD Opteron processors
358 Sun Fire X2200 M2 server compute nodes with 16 gigabytes of shared memory per node; 2 dual core 2.6 GHz AMD Opteron processors per node
55 Sun Fire X4600 compute nodes with 64 gigabytes of shared memory per node; 8 dual core 2.6 GHz AMD Opteron Processors per node
InfiniBand network cards from Sun iForce Partner Voltaire for interconnecting the nodes of the cluster
6 high-density Sun Fire x4500 data servers to collect and store information on locally shared disk managed by Lustre and available to compute nodes due to InfiniBand connection
68 terabytes Lustre scalable file system.

To collect and store the terabytes of data associated with the large HPC systems, ARSC uses Sun Fire E6800 servers backed by a new Sun StorageTek SL8500 modular library system. Specifically, the storage archiving solution is configured with: 2 Sun Fire E6800 servers, each with 8 UltraSPARC CPUs, 16 gigabytes memory, 20 terabytes of Sun StorageTek FlexLine FLX380 storage systems and 1 Sun StorageTek SL8500 modular library system with 9 Sun StorageTek T10000 and 9 Sun StorageTek T9940B tape drives. Sun SAM-QFS is the archiving software in use. High-speed file transfers between the Midnight login nodes and the archive use QFS Linux Client software.

The Sun Customer Ready Program team worked with the system architects to validate the design power and cooling needs, procured all related third-party equipment, including the Voltaire InfiniBand networking, assembled the systems, racked, cabled and tested them. Experts from Sun Professional Services were on-site for two months during the project. They performed roles ranging from to project management to installation, implementation and final acceptance. SunSpectrum Support is on-site Monday through Friday to solve any problems.

The Sun HPC solution—the first Sun supercomputer in the HPCMP program—has solidified ARSC’s position at the forefront of the nation’s university-based supercomputing centers. In addition, the Sun Opteron cluster is creating the potential for successful, international study of Arctic systems modeling. The supercomputer’s planned, optimum computing performance at teraflop speed aids scientists to predict weather patterns that have immediate impact on the safety of aircraft and passengers.