ACADEMIA

Only Weeks After System's Delivery, Researchers Show Mettle of Defense Department's Largest Computer: "HAWK," the largest and most powerful computer in the Department of Defense (DoD), has achieved new heights of application scalability for a DoD system by running a single application across 9,000 processor cores. More significantly, the feat was achieved only a few weeks after the system was delivered. An SGI Altix 4700 system from SGI, HAWK was successfully deployed ahead of schedule at the Aeronautical Systems Center Major Shared Resource Center (ASC MSRC) at Wright-Patterson Air Force Base. The supercomputer is powered by 9,216 Intel Itanium 2 processor cores and features 20TB of globally addressable memory and 440TB of usable disk space. The application-scaling breakthrough was achieved using advanced software codes that study the characteristics of turbulence. Dr. George Vahala, professor of physics at The College of William and Mary, ran a combination of Lattice Boltzmann and quantum turbulence codes on HAWK over a period of several weeks. A number of Vahala's runs utilized all available 9,000 cores. Vahala's work, which includes the study of turbulence that occurs at very low temperatures, could have several real-world implications, such as helping U.S. Government officials detect the presence of the underground tunnels used to illegally cross the U.S.-Mexico border. Through the DoD's Capability Application Project (CAP) program, more than a dozen researchers vied for the remarkable DoD's 'HAWK' Takes Off On Schedule; SGI Altix System Runs Applications on 9,000 Cores in Acceptance Tests/2 opportunity to run their applications on the Hawk system before it was commissioned for full-time production service on Oct. 1. HAWK now will serve so many researchers and scientists that single jobs will be typically limited to a maximum of 2,000 cores. "We were extremely fortunate to have this kind of dedicated access to this powerful SGI Altix system at ASC MSRC," said Vahala, whose longest single job ran just under 25 hours and represented 224,663 processor hours. Reliably scaling codes to these new heights can enable significant breakthroughs. For instance, Vahala said he now can study quantum effects of matter at resolutions never before attempted: "With 9,000 cores, we have quite easily achieved Bose-Einstein condensate grids with a resolution of 2,4003, whereas normally, the most I've seen is 5123." Under the CAP program, three other projects achieved significant scalability targets on the HAWK system. These projects, which successfully scaled above 4,000 cores, include:

• U.S. Navy studies using advanced ocean modeling codes to better predict weather events that may impact Naval operations,
• U.S. Air Force turbine studies modeling airflow through turbine engines, which may lead to smaller, lighter, and more efficient engines, and
• Georgetown University materials research aimed at developing next-generation materials for submarines and other military vehicles, ensuring they are able to outperform and outmaneuver other vehicles they may encounter.

"We are pleased to announce the newest addition to the Aeronautical Systems Center Major Shared Resource Center: an SGI Altix 4700 supercomputer with more than 9,200 processor cores," said Steve Wourms, Director, ASC MSRC. "HAWK is the largest supercomputer within the Department of Defense. We are proud to have put HAWK into service ahead of schedule, and we're excited to furnish this tool to the Department's scientists and engineers." "Increasingly, our customers are saying that they can no longer wait six months to a year before their new HPC systems are commissioned for production use," said Dr. Eng Lim Goh, senior vice president and chief technology officer, SGI. "Given the pace of Moore's Law, such delays can sacrifice a significant portion, sometimes as much as a quarter, of an HPC system's lifecycle. More importantly, these delays have also been known to adversely affect an agency's time to produce results, consequently decreasing the agency's capabilities or competitiveness." Added Goh, "The fact that the massive HAWK supercomputer was achieving scalability breakthroughs in a matter of weeks shows how SGI Altix system combine rapid deployment with productive HPC performance and scalability. Moreover, its globally addressable memory, across all 9,216 cores of the system, allows scientists and engineers to process massive problems as a whole, which delivers results faster and allows larger-scale work. SGI is proud to join with ASC MSRC as it commissions this powerful new system to full-time service ahead of schedule." Although many of today's largest supercomputers have thousands of processor cores, systems that can efficiently run an application across more than a few hundred cores are DoD's 'HAWK' Takes Off On Schedule; SGI Altix System Runs Applications on 9,000 Cores in Acceptance Tests/3 rare. The DoD's HAWK system, however, relies on SGI's NUMAlink interconnect, an extremely fast, low-latency fabric that connects HAWK's 18 SGI Altix nodes. Because NUMAlink is faster and more efficient than other interconnects, highly scalable codes, such as the four that ran on HAWK, experience low latency that impacts application performance, even when running across thousands of processors. More than a thousand researchers will use HAWK to design weapons systems faster, reduce risk by increasing the quality of modeling and simulation, and support an intensifying effort to develop "game-changing" computational science and engineering applications.