In less than two months, the National Strategic Computing Initiative (NSCI) Executive Council must present its implementation plan. Just what that will look like remains a mystery but budgets, governance, and private-public partnering models were on the minds of attendees to last week’s HPC User Forum in Broomfield, CO, where the first public panels with NSCI agencies offered a wide-ranging glimpse into agency thinking. It was also a chance for HPC industry execs to press for more details.
At the moment industry enthusiasm for NCSI is high and panelists strove to reinforce that goodwill and reassure attendees that disabling missteps could be avoided. On technology issues – despite differences around the edges – there was broad agreement among the panelists and attendees on problems needing solving, with large-scale data analytics as the new but important kid on the block. Governance and collaboration challenges drew a more wary response from the audience, although panelists insisted conflicts would be amicably and equitably dealt with.
In the end conversation around non-technology issues was the most revealing with many attendees wondering what potential obstacles the NSCI panelists foresaw. Funding, perhaps not surprisingly, was a touchy topic particularly given the many calls that NSCI should emulate the U.S. Apollo program, which galvanized public opinion and loosened federal purse strings.
Playing devil’s advocate, Barry Bolding, Cray (NASDAQ: CRAY) senior vice president and chief strategy officer, said to panelists during Q&A, “I’d like to push the panel in the direction of pitfalls a little bit and hear about things you think could be gotchas. You’ve mentioned the space program a few times and how this might be a corollary and one can look at the space program and all of us agree it benefited the country a great deal. [But] one can also look at it and say, oh it’s been 45 years and we’re only beginning to have a vision for unmanned space program and only beginning to get private industry into space programs. So it wasn’t very successful.”
Put another way, what did we get for the money and given the times, can we get those kinds of budgets going forward.
This drew a pragmatic response from OSTP representative Randy Bryant: “It’s hard to get significant federal funding output [now]. Flat is the new normal; that’s true across the entire research budget and I see that as a core problem. The Apollo program was a great program but it consumed a significant fraction of the U.S. GDP. It was a huge investment, and I don’t anticipate in our current budget climate that would be possible,” he said. “It helped that there was an existential threat of the Soviet Union at the time. I don’t see anything that’s going to make us step up at that level.”
Rob Leland, one of the original organizers of the NSCI proposal and a representative from Sandia National Laboratories, countered by saying the real gotcha is to not sufficiently fund and execute NSCI.
“The US used to dominate investment in this space quite dramatically. In fact up until about 2010 or so U.S. investment was equal to the rest of the world combined and [it’s] now about a 1/3 of the total investment. More worrying is the disparity in growth rates. The U.S. growth rate in investment is about 2.5% [while] the average for the rest of the world that is engaged in this space is about 12 % and I think China is up to about 23%.
“If that disparity persists for five or 10 years we will not dominate this space technologically the way we have previously,” Leland said.
If that wasn’t compelling enough, Leland emphasized the “erosion of Moore’s Law” has upped the ante in HPC competition. “If we don’t rally effectively as a society around that challenge the technical path forward is very unclear. I think there are also good indicators we’re [also] coming to the end of the MPP era and so if we don’t make a transition to some new architecture approach, I think we will be on a path of less relevance.”
In many ways, the event marked the beginning of NSCI’s public outreach to industry in a program, which among other things, is designed to energize public-private partnering for the good of both. There were two panels: 1) US Plans for Advancing HPC: Potential Implications of the White House Executive Order and NSCI, and 2) Open Forum Discussion and Q&A of the NSCI Plans and Directions.
The roster of panelists was impressive: Bryant, OSTP; Irene Qualters, National Science Foundation; Doug Kothe, Oak Ridge National Laboratory; Will Koella, DOD/NASA; Bert Still, Lawrence Livermore National Laboratory; Piyush Mehrotra, NASA Advanced Supercomputing Division; Bill Kramer, NCSA; Nathan Baker, Pacific Northwest National Laboratory; and Leland, Sandia. Bob Sorenson of IDC moderated both panels.
As a rule, panelists directed their comments to one or another of NSCI’s five strategic objectives excerpted from the Executive Order here:
- 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 analytic 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.
Not surprisingly panelists’ comments largely reflected their specific agency missions – this was helpful in making clear there are some differing priorities. To a large extent the list of technology issues tackled was very familiar to anyone in HPC: the end of Moore’s Law; death of single thread performance; power management; a need for higher fidelity models; the flood of data from scientific and other instruments and sensors; code modernization; and future computing (quantum, neuromorphic, et. al). You get the picture.
The problems are plentiful and solutions scarce, but that’s kind of the point of NSCI. Big Data and code modernization took somewhat center stage. Here are a few examples:
- Kothe (ORNL) pushed the importance of codesign citing ongoing work: “In DOE there are three of those centers and I know the NNSA labs are heavily involved in that activity. In the ECI (Exascale Computing Initiative, DOE, and being somewhat subsumed into NSCI) project we see that activity continuing and growing. It’s critical because we are looking at some fairly substantial challenges at least on the applications side, the scariest are the deep memory hierarchies, probably more so than hybrid floating point.”
- Mehrotra (NASA) locked in on big data issues: “We are very interested in the convergence of data analytics with HPC. Our satellites produce petabytes of data every year streaming down. This is observational data. How do we handle that data, how do we manage that data, and then how do we actually extract any knowledge out of that in conjunction with not just observational data [but with] model data. We are very concerned about how to bring the two environments together so that we can do quantitative simulation along with large-scale data analytics.”
- Still (LLNL) sounded a familiar note on code modernization, “You’ve heard the acronym IC used for Intelligence Community, we used it in a slightly different way for integrated code. The gist is our IC [effort] is multi-million lines of work across the three labs within the NNSA; it’s kind of a $6B investment in code. We can’t rewrite it overnight and take advantage of each new architecture that shows up because the codes are decade-old type codes. We have to make modifications or reengineer one and revalidate. So performance portability is an absolute key inside the NCSI. We are all about trying to make useable machines. That is a key component as far as we’re concerned.”
- Baker (PNNL) offered: “The amount of data that comes off a big instrument is too high a bandwidth to even write out to a box. So you’ve got a ‘baby and bathwater’ conundrum. We spend billion of dollars looking for rare particles and yet the data is coming out at a rate that we may have to triage, we may lose what you’re looking for. How do you design robust algorithms that can handle that? How do you design algorithms that can detect what you need to detect and although you’d love to keep all the data, triage what you have to triage?”
Transferring NSCI-generated technology advances to industry got perhaps less shrift than one would think. Kramer (NCSA) strongly suggested that HPC-as-a-service must be a necessary component of any realistic approach to induce widespread use of HPC technology by most of industry.
“We’re talking about the management of IP and partnerships and relationships very seriously,” said Koethe (ORNL) “and that scope is probably not as deep and broad as it should be. I neglected to show our structure which call for councils – industry council, science council, board of directors. Not that boxology fixes everything but I think at least we’re implementing lessons learned and best practices from past projects.”
Attendees dedicated a fair amount of discussion to coordination challenges within the NSCI organization framework. Competition among government agencies for power and budget is hardly rare. Diverse suggestions ranging from close coordination, loose coordination, sole agency lead, collective agency lead and others were all raised at some point.
Qualters (NSF) said simply, “I think this is a very aggressive program and there’s not one path forward. I think one has to be careful. One wants a fair amount of innovation at this stage and diversity. So there can be coordination but I think the [idea] that you just have everyone marching in one line is wrong too.”
What came through is a hunger for a good model for success. This is a sprawling program – and those have been tackled before (e.g. Large Hadron Collider) and produced many lessons. One audience member looked back to the rural electrification project in the late 1930s as a good model, particularly from a public-private partnership perspective. It was long-lived and worked. The HPC initiative (High Performance Computing Act 1991) in the early 90s seemed to be the most favored.
As an early architect of the NSCI directive, Leland offered this: “I think there is an excellent analog in the HPC initiative in the early 1990s that is generally viewed as quite successful and I think we can hope to replicate that success. [If] you look at history I think each major new era in computing has been preceded by 5-7 years by a forward looking investment by the government in R&D. [The pattern can be] traced that back at least five cycles. I think that can be true again here. There are many indicators that we are approaching a wall and need to make a substantial jump in our capabilities and a change in our approach. I think all the preconditions are here for us to replicate that history once again with a sixth cycle.”
It will be interesting to see how opinions shift once the implementation plan comes out.