Last Friday, almost exactly one year after President Obama’s Executive Order establishing the National Strategic Computing Initiative (NSCI), the Office of Science and Technology Policy held a workshop in Washington D.C. ostensibly to celebrate NSCI accomplishments so far – and there are many. Just as important, the meeting sought to (re)energize NSCI’s diverse stakeholders, many of whom were early enthusiastic supporters but have become wary because of a perceived lack of action to implement NSCI.
On balance, attendees seemed encouraged after the meeting. The sub-text was that delivering such an expansive, whole-of-government plan during the current fractious election year is an enormous challenge and that patience in addition to forward movement is required. Acknowledging the lame duck status of the current administration, Jason Miller, Deputy Assistant to the President and Deputy Director of the National Economic Council, and the closing speaker, emphasized, “This administration only has six months left and there’s one thing we are focused on every day; [it’s] what we call hardwiring. How do we hardwire these changes in.”
The workshop was an invitation-only event held at the Eisenhower Executive Office Building next to the White House. It drew about 85 attendees from industry, academia, and government. The meeting also coincided with release of the new NSCI Strategic Plan, presumably the renamed “implementation plan” mandated in the initial executive order.
Here are two comments typical of the generally positive tone of attendees present:
- Barry Bolding, chief strategy officer of Cray said he was impressed “with the diversity of the group that OSTP brought together. Specifically with regards to the broadening the public-private collaborations it was good to have participation from the policy side with Chris Mustain from the Council on Competitiveness as well as major technology and system vendors. This broad discussion is essential for Public-private collaborations needed to broadly reach across various applications areas to ensure our strategic computing initiative targets US competitiveness in both computer architectures and productivity of these systems.”
Ilkay Altintas, chief data science officer, San Diego Supercomputer Center, said “I thought the workshop was good to remind everyone that the NSCI is not just important, but also vital to the progress of science and future US innovations in computing…It will require innovation within the current platforms and increased fundamental research in new computing paradigms. It was also great that there was a lot of discussion on workforce development and convergence of big data analytics and traditional supercomputing. I tried to make the point that software can be used to lower the barriers to computing and also improve the capacity and capabilities of the current HPC ecosystem.”
The roster of speakers was drawn from across NSCI constituencies and each gave brief, roughly five-minute talks. Breakout sessions at each table tackled specific NSCI topics with an eye towards suggesting next steps for operationalizing NSCI. There was little downtime. While technology infrastructure (hardware and software) received affair amount of discussion, it was the workforce challenge – the HPC workforce is too small with no obvious way to grow it at scale – and the need for improved public-private R&D partnerships that seemed to receive the most attention.
Speakers included: Tom Kalil, deputy director, OSTP; Deborah Frincke, director of research, NSA; Al Gara, fellow and chief architect for Exascale Systems, Intel; Altintas, chief data science officer, San Diego Supercomputer Center; Jim Kurose, assistant director for Computer and Information Science, NSF; Michael Rosenfield, VP for Data Centric Solutions, IBM Research; Henry Neeman, director, University of Oklahoma Supercomputing Center; Lynn Orr, Undersecretary for Science and Energy, DOE; Marti Head, director of ‘The Noldor’, GlaxoSmithKline; Ivet Bahar, distinguished professor, University of Pittsburgh; and Miller.
By and large, the right people seemed to be in the room although one attendee wondered, “Where were some of the other NSCI agencies, especially the deployment agencies like DHS, FBI, NOAA, and NIH? They have been relatively silent in the process so far, and I wonder what that means?”
Before plunging into workshop discussion topics – most of which are very familiar (workforce, Moore’s Law, energy requirements, converged data analytics-traditional HPC infrastructure, private-public R&D partnerships, et al) – it is worth noting there’s was agreement from the Obama administration that more federal R&D spending is needed. The U.S. Government, said Miller, has greatly underinvested in R&D while industry has done a strong job particularly in the post 2008 Great Recession recovery.
“If you look right now as a country (including business), we’re investing almost 3 percent of GDP in R&D. We’ve seen real improvement from the private sector in the last several years. When you look at federal investment in R&D, it’s about 0.8 percent of GDP, near historic lows. We can’t continue to underfund federal R&D and expect to drive innovation for years into the future,” said Miller in wrapping up the workshop. R&D spending is the seed corn for innovation, he said, and that seed corn is at risk.
Miller’s remarks followed earlier comments by Kruse of NSF, who likewise emphasized that NSF’s first-year $30 million invested by NSF was just the start. New monies from various quarters should be expected was the underlying suggestion and he rather dramatically said he sleeps with a copy of the NSCI Executive Order beside his bed – it’s that important. It will be important to watch how effectively increased funding intentions turn into actual budgets.
There was not a lot of discussion of China’s ascent in the HPC world but certainly the awareness was there. One attendee said, “The presentation agenda and content reinforced that a comprehensive and long-term commitment is needed. The NSCI was prescient in light of the last year’s events.” So it was referring to China placing the two top machines at the top of the Top500 List and pulling even with the U.S. in terms of the number of machines placed on the Top500. On that note it may be worth quickly skimming NSCI’s four overarching principles and five objectives bulleted below:
- The United States must deploy and apply new HPC technologies broadly for economic competitiveness and scientific discovery.
- The United States must foster public-private collaboration, relying on the respective strengths of government, industry, and academia to maximize the benefits of HPC.
- The United States must adopt a whole-of-government approach that draws upon the strengths of and seeks cooperation among all executive departments and agencies with significant expertise or equities in HPC while also collaborating with industry and academia.
- The United States must develop a comprehensive technical and scientific approach to transition HPC research on hardware, system software, development tools, and applications efficiently into development and, ultimately, operations.
- Accelerating delivery of a capable exascale computing system that integrates hardware and software capability to deliver approximately 100 times the performance of current 10 petaflop systems across a range of applications representing government needs.
- Increasing coherence between the technology base used for modeling and simulation and that used for data analytics computing.
- Establishing, over the next 15 years, a viable path forward for future HPC systems even after the limits of current semiconductor technology are reached (the “post- Moore’s Law era”).
- Increasing the capacity and capability of an enduring national HPC ecosystem by employing a holistic approach that addresses relevant factors such as networking technology, workflow, downward scaling, foundational algorithms and software, accessibility, and workforce development.
- Developing an enduring public-private collaboration to ensure that the benefits of the research and development advances are, to the greatest extent, shared between the United States Government and industrial and academic sectors.
Without question, NSCI represents optimistic, expansive thinking on a grand scale. It was greeted by HPC community-wide enthusiasm early on. But the follow-through has been, perhaps predictably, challenging for such a large whole of government program.
Many projects cited at the workshop – the Cancer Moonshot, for example – were presented as examples of current success and representative of future activities. A few observers contended some of the cited examples were really pre-existing programs simply being subsumed under NSCI. The exacale computing program, for example, was already well framed and under way. The idea here is that those programs whose goals seem aligned with NSCI are now being co-opted under its umbrella. Maybe that’s not a bad thing, argued others, if they are not the only things that NSCI does.
Al Gara, fellow and chief architect of exascale systems, Intel, gave a solid presentation emphasizing the need for long-term pubic-private partnerships on research over multiple generations. In a four-year program, for example, he said, there are so many technology unknowns “from the silicon all the way to the architecture, with innovation going on. We have to figure out how to do this together as opposed to this very segmented way we currently do.” This isn’t a new problem and perhaps NSCI can provide new frameworks for tacking partnerships.
Irene Qualters, division director for Advanced Cyberinfrastructure at NSF, echoed the need for creative thinking around collaboration. “Discussions from the event made it clear that the NSCI will require new forms of partnerships across industry, academia and government in order to research, test and prove successful HPC approaches. [At the meeting] there was early evidence of greater interagency coordination as well as initial engagement with academia and industry,” said Qualters.
Gara also made a point of toeing the “Moore’s Law is not dead” line while assuring all that Intel is well invested in looking ahead at the usual crop of emerging technologies. That’s in addition to exploring how to extend silicon, for example, by moving into 3 dimensions. The reality seems that near-term at least, CMOS will remains the workhorse for general purpose while other technologies potentially carve out niches.
Perhaps not surprisingly, there was general agreement about the importance and rapid rise of data driven science and the need for converged or at least co-capable architectures. Ilkay of UCSD discussed the WiFIRE program she leads. She’s a champion of workflows and virtualization techniques. “We built this scalable infrastructure that can be utilized for dynamic, data-driven decision support for wild fire prediction. The approach here is to take advantage of real time data, learn about the dynamics of the situation, and then be able to assimilate that into the fire models. Workflows here are used for system integration. I am hoping the NSF systems of the future will benefit from the hybrid technologies.”
IBM’s Rosenfield reviewed Big Blue’s efforts in neuromorphic and quantum computing, both of which may become important in the future. “We are trying to build a universal quantum computer, not a specialized system, but a universal system that you can use on multiple types of applications,” he said, adding the forty thousand users and logged on and taken advantage of IBM 5-qubit systems that available in the cloud (http://www.research.ibm.com/quantum/).
On one of the thorniest issues, workforce needs, Warren Kibbe, director of that National Cancer Institute’s Center for Biomedical Informatics and Information Technology, NIH, and a key leader of the Cancer Moonshot program, said, “We need to improve opportunities for formal and informal training in HPC, and promoting diversity in the workforce will be crucial for continued success. The application of HPC in different sectors also requires a trained workforce, and individuals with cross-disciplinary training. Developing additional training opportunities was a clear message we heard today.”
Indeed there were many conversations around workforce development and several ideas for traditional academic programs and informal programs, the latter targeted at people already in the workforce. Neeman of the University of Oklahoma focused on workforce issues in his talk. He called for a show of hands from those whose table discussion focused hardware and software versus soft issues including workforce and the latter dwarfed the former.
“Teaching HPC is hard. It’s not the individual skills; it’s the stack of skills that you need, the combination, in order to be good at this and the number who have that combination is far lower than we need,” Neeman said. While most talks drew very few questions, his sparked several. It seems likely there won’t be a single solution but several targeted to different groups and different expertise levels. One study, he noted, indicated it takes two years from the time of hire to train HPC staff – not an especially attractive scenario for companies.
All of the short presentations were fascinating and intended to provide glimpse into the kinds of problems HPC can help solve and showcase a few that fall under NSCI auspices. DOE actually provided lengthy scorecard and playbook for NSCI activities in a blog on Friday, parts of which are excerpted below. Though a bit lengthy, it’s worth scanning.
One of the most thoughtful assessments of the meeting was provided by Bob Sorensen, VP of Research, IDC HPC group. Sorensen spent 30-plus years in government as an HPC analyst and was there during the early formation of the NSCI.
“Based on the discussions I had and heard [today], my sense is that everyone now ‘gets’ the NSCI, but that everyone essentially resonates on the aspect that they are most comfortable with: some talked about post-CMOS, others about exascale systems, others about software workflow or workforce creation. Indeed, one or two even talked about economic competiveness. It’s clear that the complexity and ambition of this program are high, and the need to manage all of the moving parts is critical, lest some get neglected.
“This seemed to be the case in the presentations as many showed current achievements and plans for future efforts, but did little to place them in context with the overall vision of the program (Most, I suspect, would have been the same presentations regardless of the existence of the NSCI anyway.) This is to be expected, I suppose, as there really is no central narrative that everyone can rally around. Indeed, previous ‘spokesmen’ for the program, guys like Rob Leland, Randy Bryan, and Saul Gonzalez, have already moved on to other jobs so there is no real focal point for the effort with much continuity. Left without such, my sense is that each agency will go off and do its thing based on its own interpretation of the NSCI. And that means that the advantages of a ‘whole of government’ effort could be lost.
“This also, to my ear, led to some confusion from the industry players at the workshop. Simply put, if they want to get involved, what is their next step? (i.e. Who do I call?) A strategic plan that says things like “You can respond to RFIs” does little to help those who want to play, and does nothing for those that should be playing but don’t even know where to start (like most of US industry that is not HPC savvy, but who could benefit from HPC in their R&D or business process).”
The workshop was rich in content and ideas. It did seem to reassure several attendees that the NSCI flame is being carried forward in earnest by the Obama administration during its last months. How successful the effort will be remains to be seen. HPCwire will take a more focused look at the new NSCI strategy plan soon. Excerpted directly below is DOE’s list of present and planned NSCI activities.
EXCERPT FROM DOE RELEASE ON NSCI ACCOMPLISHMENTS AND PLANS
“In addition to the coordinated interagency efforts, Federal agencies have also undertaken a number of individual actions in support of the NSCI objectives. For example;
The Department of Energy (DOE) has:
- Invested in a suite of efforts to continue developing the research base for “Post-Moore’s Law era” computing. These activities include convenings, such as a workshop on neuromorphic computing and a science roundtable with representatives from national laboratories and universities on quantum-based sensors; release of a request for proposals on machine learning research; and creation of strategic partnerships with the private sector to establish testbeds for Post-Moore’s Law computing systems in energy-efficient machines that mimic the brain’s abilities (including a D-Wave 2X system deployed at Los Alamos National Laboratory and a 4 billion synapse neuromorphic system hosted at Lawrence Livermore National Laboratory).
- Established a series of NSCI interagency and public-private collaborations that also support other Administration Initiatives. DOE is supporting the Vice President’s Cancer Moonshot in several ways, including the creation of the Joint Design of Advanced Computing Solutions for Cancer program at the National Cancer Institute (NCI) to accelerate progress in pushing both the frontiers of high-performance computing and predictive oncology and the launch of a public-private partnership to accelerate precision medicine from molecule to patient that initially includes the NCI and GlaxoSmithKline but hopes to include additional partners. It has also established a collaboration with the Department of Veterans Affairs (VA) to more broadly support the Precision Medicine Initiative by applying advanced supercomputing capabilities and data analysis to the nearly half a million veterans’ records from one of the world’s largest health research cohorts, the Million Veteran Program (MVP). This five year partnership is now underway and the first phase will focus on cardiovascular health, mental health issues, and prostate cancer.
The National Science Foundation (NSF) has:
- Issued the Energy-Efficient Computing: from Devices to Architectures (E2CDA) program solicitation with the private sector to support new research to minimize the energy impacts of processing, storing, and moving data within future computing systems.
- Deployed four new national data-focused HPC resources in the last 14 months, the largest number and the most diverse set of NSF-supported HPC resources ever deployed in a similar time period, which includes Jetstream at Indiana University, Bridges at the Pittsburgh Supercomputing Center, Comet at the San Diego Supercomputing Center/University of California San Diego, and Wrangler at the Texas Advanced Computing Center/University of Texas at Austin.
- Issued a Software Institute Conceptualization award that sponsors community workshops and conceptual work to take advantage of the significant data and computing requirements of the Large Hadron Collider as a science driver for next generation high-performance software and sustainability developments.
The Department of Defense (DOD) has:
- Initiated the Quantum Science and Engineering Program in the Office of the Assistant Secretary of Defense for Research and Engineering, focused on accelerating critical technologies for quantum networks and sensors that enable capabilities impossible to achieve with classical systems over the next three years.
- Implemented a micro-services architecture for the Army’s High Performance Computing Modernization Program Portal, a framework for building easier-to-use HPC applications that provides zero-footprint secure access to high performance computers via a web browser, to enhance performance for the current user base and be affordably scaled to a larger user base.
The Intelligence Advanced Research Projects Activity has:
- Launched two new funding programs focused on advancing HPC hardware, software, and capabilities. The Logical Qubit (LogiQ) program aims to overcome the limitations of current multi-qubit systems by building a logical qubit from a number of imperfect physical qubits, and the Machine Intelligence from Cortical Networks (MICrONS) program seeks to revolutionize machine learning by reverse-engineering the algorithms of the brain.
- Published two new Broad Agency Announcements for the Quantum Enhanced Optimization (QEO) program, which seeks to harness quantum effects required to solve complex problems, and the SuperTools program which will develop a new superconducting circuit design flow.
- Released RFIs on Energy Efficient Computing Architectures, Data Ingress and Data Egress in Cryogenic Systems, Neurally Inspired Computing Principles, Electronic Design Automation tools for Superconducting Electronics, and Spectral Holographic Optical Processing.
The National Institute of Standards and Technology (NIST) has:
- Announced the HPC Security Workshop, to take place in Fall 2016, to identify security priorities and principles that should be incorporated into NSCI and bring together stakeholders from industry, academia, and Government, and to identify gaps that should be addressed.
- Created an NSCI seminar series to cover a broad array of topics designed to increase the knowledge base of NIST staff in advanced computing and data science as well as highlighting specific areas where NIST is likely to play a critical role, such as materials for future computational platforms, measurement and testing science, algorithms and applied mathematics, and neuromorphic and quantum computing.
NEW ACTIONS ANNOUNCED TODAY
In addition to the activities over the last year described above, Federal agencies are announcing a new set of actions today that will further the objectives of the Initiative and engage the academic and industry sectors in support of NSCI. These steps include:
- NSF is establishing two multi-institutional Scientific Software Innovation Institutes that will develop sustainable community software platforms in support of NSCI. One institute, based at the University of California, San Diego will focus on supporting the development of advanced portals to democratize HPC access. The other institute at Virginia Tech will develop an end-to-end cyber ecosystem for the molecular sciences community.
- Today, the University of Maryland UMD and the U.S. Army Research Laboratory (ARL) announced a strategic partnership that enables the broad deployment of high-performance computing (HPC) resources and creates a national model of research collaboration. As part of ARL’s Open Campus Initiative, and in collaboration with DOD HPC Modernization Office, an ARL supercomputer will be made available to academia and private sector innovators through UMD’s Mid-Atlantic Crossroads (MAX) cyberinfrastructure for application development and networking experiments.
- Today, Federal agencies released a white paper describing the interagency vision for the emerging and innovative solutions needed to realize the Nanotechnology-Inspired Grand Challenge for Future Computing announced in October 2015. The white paper, a collaboration by the DOE, NSF, DOD, NIST, and the Intelligence Community, describes the interagency technical priorities, highlights the challenges and opportunities associated with these priorities, and presents a guiding vision for the research and development needed to achieve key technical goals for the challenge.
- DOE is announcing advances made in the design and development of its Exascale Computing Project (ECP). Achievement of this milestone is based on a year’s work following the signing of the EO, including formal inauguration of the ECP, calls for proposals for exascale hardware and software technologies, and identification of the first round of ECP applications software areas.