NCSA's Bill Kramer Assesses NERSC's HPC Legacy

At the Feb. 4-5 NERSC User Group meeting marking the center’s 40^th anniversary, invited speaker Bill Kramer used the occasion to assess NERSC’s legacy to the high performance computing (HPC) community. Kramer, who is the director and principle investigator of the Blue Waters Project at the National Center for Supercomputing Applications in Illinois, was one of the first NERSC employees hired by LBNL when the center moved to Berkeley Lab. He spent 12 years as NERSC’s general manager.

During his time at NERSC, Kramer was one of the drivers of LBNL’s Science Driven Computer Architecture initiative and consistently emphasized that systems be evaluated on their performance of real-world applications, not theoretical peak speeds.

Before joining NERSC, he worked at the NASA Ames Research Center, where he was responsible for all aspects of operations and customer support for NASA’s Numerical Aerodynamic Simulator (NAS) supercomputer center.

During his career, Kramer has led the acquisition, testing and introduction of more than 20 high-performance computing and storage systems. He was named one of HPCWire's "People to Watch" in 2005 and 2012, InsideHPC’s “Rockstars of HPC” in 2010 and he chaired the SC05 conference.  Now he is leading the National Science Foundation’s investment in HPC: the Blue Waters sustained petascale project (bluewaters.ncsa.illinois.edu).

Although Kramer said that his views were his own, he based them on his experience working at three world-class computing organizations, as well as his time in academia and industry. For his talk, Kramer defined NERSC as comprising the clients and users, philosophy, staff, resources, funders, partners, associates, collaborators and friends of the organization.

“NERSC is the first, multi-discipline, nationally based HPC Facility for diverse, open science,” Kramer said. “And NERSC taught the rest of HPC community how to do that type of computing.”

Since 1974, when it was created to serve the fusion energy community, the center has had direct influences on computing centers serving NASA, the National Science Foundation, the National Center for Atmospheric Research, the Department of Defense, DOE’s Accelerated Strategic Computing Initiative and Leadership Computing Facilities program, as well as international HPC centers Kramer said.

Kramer gave several examples of how NERSC has influenced the HPC community.

Focus on science first: NERSC has made exceptional customer service an integral part of its mission, which leads to scientific productivity. Kramer noted that “a lot of hard work and innovation goes into this excellent customer service,” and many of these methods have been adopted by other centers.

NERSC and its partners helped create new software capabilities: Over the years, NERSC staff have worked with developers to make production-quality enhancements to HPC, visualization tools, data management software, adaptive load-balancing resource management technologies, I/O and storage, programming models (such as UPC), cyber protection that does not interfere with users, and applications in areas ranging from astrophysics and climate to materials.

NERSC merged Big Compute and Big Data decades before we knew to call it Big Data: NERSC was one of the original partners in the National Storage Laboratory, which in turn led to the development of HPSS, the High Performance Storage System. NERSC was also the first center to deploy the Global Unified Parallel File System, which allowed users to access their files across all systems. Other centers tried and failed at global file systems, saying it could not be done. NERSC showed it was possible and the approach was emulated elsewhere. NERSC was also the first center to merge HPC simulation computing with HPC data analytics of observational data in an organized manner, an approach that helped Saul Perlmutter win the 2011 Nobel Prize in Physics.

NERSC focuses on real-time, real-world performance: The system acquisition process of systems at NERSC has always been driven by researcher’s science needs and shortening the time-to-solution for real problems. The Sustained System Performance (SSP) benchmark developed at NERSC for procurements has been adopted by a number of other centers.

NERSC is the best at showing its real impact on science: Many centers struggle with communicating their worth, Kramer said, but NERSC does an outstanding job of highlighting the number of peer-reviewed papers written by users, the number of times those papers are cited, the uniqueness of the results and especially the unexpected findings.

NERSC and LBNL helped reinvigorate federal funding for supercomputing, particularly for open science: Kramer said it was a little-known story, but in 2004 Berkeley Lab Director Chuck Shank invited his friend Ray Orbach to the lab for a meeting to discuss funding for HPC. At the time, Orbach had not yet been confirmed as the head of DOE’s Office of Science. Shank, Bill McCurdy, Horst Simon and Kramer spent half a day explaining the issues, the importance of and motivation for increased federal investment in HPC. Since then, Kramer noted, the federal budget for HPC has doubled and the portion for open computational science has more than doubled, a remarkable accomplishment since the HPC budget had previously been dominated by classified work.

NERSC is entirely open and on the Internet: NERSC and ESnet have helped create and implement new cyber protection methods, including the Bro intrusion detection system (now being adopted by NSF centers among others) and the Science DMZ approach for improving the flow of scientific data.

NERSC provides persistent support for all types of computing: The DOE INCITE (Innovative & Novel Computational Impact on Theory and Experiment) program was launched at NERSC in 2003 to serve large-scale computing projects led by well-qualified teams. After proving its scientific worth, INCITE was expanded to other DOE sites. But Kramer pointed out that the “missing middle,” those mid-range computing projects, have no better place to work than NERSC.

Along the way, NERSC also helped reinvigorate scientific computing at Berkeley Lab, which in turn helped the lab attract new, younger, more computationally focused researchers.

In closing, Kramer summarized NERSC’s legacy:

• NERSC was the first.
• NERSC is there the longest.
• NERSC is open science at unprecedented scales and impact.
• NERSC is a perennial leader of the supercomputing community.

“The nation, DOE and the supercomputing community should not only be proud of NERSC, but also thankful for NERSC,” he concluded.