In this special guest feature, John Kirkley describes how industry thought leaders are preparing for the National Strategic Computing Initiative to enable exascale computing.
How will the National Strategic Computing Initiative (NSCI) move the U.S. forward toward exascale computing?
Examining that question was the focus of a spirited “NSCI and the Democratization of HPC,” panel discussion—now available on video—held at Intel’s Community Hub during last November’s SC15 show.
NSCI was created by a White House executive order in July 2015. Designed to ensure that the United States remains a leader in HPC over the coming decades, the initiative focuses on collaboration between government, industry and academia in order to achieve the Holy Grail of exascale computing in the not too distant future.
According to the White House, “This coordinated research, development, and deployment strategy will draw on the strengths of departments and agencies to move the Federal government into a position that sharpens, develops, and streamlines a wide range of new 21st century applications. It is designed to advance core technologies to solve difficult computational problems and foster increased use of the new capabilities in the public and private sectors.”
The panel, chaired by Mike Bernhardt, Community Evangelist / HPC & Technical Computing Segment at Intel Corporation, included: Tim Polk, Assistant Director for Cybersecurity at the Office of Science and Technology Policy; John West, Director of Strategic Initiatives for TACC (Texas Advanced Computing Center); and Rudolf Eigenmann, program director with the National Science Foundation.
The panelists quickly moved into a discussion of how to precisely define NSCI’s charter and how their respective organizations might contribute. For example, West, who had spent years focused on HPC for the Department of Defense, said that in his current position with TACC, he and his team can concentrate on what needs to be done to build an “exascale ecosystem” that incorporates all the elements required to create and field these highly advanced systems.
To get there we need innovation not only in hardware and programming, but when we put an exascale system on the floor we must be able to run it effectively,” he said.
This means attending to a myriad of mundane but important tasks such as systems management, code modification and continuous performance monitoring – absolute essentials when dealing with systems of this size and complexity. He added that we will need to thoroughly understand what is happening inside the system at all times in order to empower users to take full advantage of the highly advanced capabilities of these powerful supercomputers.
Computer scientists involved in the move to exascale will intervene directly with users to help them improve their code and partner with them early in the exascale design loop to address their needs – a unique perspective. The approach that West described meshes precisely with the NSCI initiative’s goals.
Polk described the efforts leading up to getting the executive order signed – a process that took two and a half years working with government agencies focused on economic competitiveness and scientific discovery. The first item on the new initiatives agenda is to build an exascale system that will provide a 100X improvement in running critical applications, which are being handled by 10 petaflop systems today. Like West, he emphasized the need to build an exascale ecosystem based on public/private collaboration.
And West, underscoring Polk’s comments, noted that government provides a “bully pulpit” that can be used to influence the direction exascale technology can take. This includes creating a national HPC infrastructure dedicated to advancing the country’s competitive position by enabling breakthrough discoveries in science and technology.
Industry, academia and government each have their own special roles to play in realizing NSCI’s objectives. Because academic institutions and corporations can be more nimble than government, their individual research and development will constitute an important part of the efforts to meet NSCI’ goals. But the primary emphasis is on cooperation between these three sectors with the government leading the way.
For example, Eigenmann said that the NSF’s key charter is research, so his agency is focused on aspects of NCSI that have research components. NSF is already engaged in a number of programs involving the private sector and these efforts will become part of drive to help meet NSCI’s objectives. The NSF initiatives include providing seed money to academia and industry, including startups, to formalize the relationship between these sectors.
HPC has matured to a point where it is now ready to be deployed widely in industry,” he said. “Business is saying that to date we have used HPC technology in a small way, but now we realize that we can make significant progress using this technology very widely. And, the research (conducted) by academia that is being fueled by NSF seed money can be transferred to industry. Our programs are designed in such a way that industry can quickly become the sole financial support of these startup programs.”
One problem in building an exascale ecology that Eigenmann touched on was workforce development. Especially given that “very few of our young people want to go into engineering and science.” NSF is making extra efforts to help alleviate this situation.
Opening up a new topic, panel chair Mike Bernhardt asked how NCSI will transcend next year’s changes of administration when the dust has settled from the presidential elections.
Responded Polk, “We expect NCSI to run for the next two decades. It’s a bit audacious to start a 20 year project in the last 18 months of an administration, but one of the things that gives us momentum is that we are not starting from a clean sheet of paper. There are many government agencies already involved and what we’re really doing is increasing their coordination and collaboration. Also we will be working very hard over the next 18 months to build momentum and establish new working relationships with academia and industry.”
West added that in the process of responding to the NSCI initiative “We can make a transformative impact on the way we do computing…we may surprise ourselves and do something fundamental that will affect future generations.”
Another reason NSCI is worth doing, he said, is that we will help create a whole cohort of undergraduate students that will have the initiative as part of their DNA. We are building an innovative generation, a fact that could pay great dividends in the future. “That’s exciting.
Computing initiatives at this scale are a team sport,” he continued. “We need everyone to contribute and bring their perspective into the mix so we can get the best solutions possible.”
West noted that during the formation of NSCI, there was a lot of discussion around the many opportunities for disciplines outside the realm of hard science and engineering – the social sciences for example. The initiative should embrace these disciplines and make exascale-class systems available to educators and researchers in academia and industry “on their own terms.” This means making these highly advanced systems accessible to users without their needing a degree in computer science – in other words, West said, the creators of these systems must go far beyond just providing a command line interface. The term “user friendly” takes on a whole new meaning in this context.
Polk summed up the NSCI this way: “NSCI is government trying to be organized and take a holistic approach to the problem (of developing exascale systems). Answers in HPC have always been most efficient when there was close collaboration between industry, government and academia. NSCI is not just a governmental, but also a national approach. We’re looking for a substantial number of enthusiastic professionals from government, academia and industry to join us as we go forward over the next few decades. This is an unprecedented opportunity to create an exascale ecosystem; an ecosystem that will become an integral part of an advanced computational fabric that will drive a new era in science and engineering.”
Watch the video covering the entire panel discussion.