June 1 — The 2015 Blue Waters Symposium, held May 10-13 at Oregon’s beautiful Sunriver Resort, brought together leaders in petascale computational science and engineering to share successes and methods.
Around 130 attendees, many of whom were Blue Waters users and the NCSA staff who support their work, enjoyed presentations on computational advances in a range of research areas—including sub-atomic physics, weather, biology, astronomy, and many others—as well as keynotes from innovative thinkers and leaders in high-performance computing. Over the three days of the symposium, 58 science teams from across the country presented on their work on Blue Waters.
“This event showcases the importance of a computational resources of this size,” said NCSA Director Ed Seidel. “Thanks to Blue Waters, detailed simulations of complex astrophysical phenomena, HIV, earthquake events, and industrial engineering processes are being done, leading to major scientific breakthroughs or new products that cannot be achieved any other way.”
Indeed, the importance of access to the computational power of the Blue Waters supercomputer could be heard from one presenter after another, no matter the research domain.
Thomas Cheatham, associate professor of medicinal chemistry and adjunct assistant professor of bioengineering at the University of Utah, is working on accurate modeling of RNA and other biomolecules, like nucleic acids and proteins. Thanks to Blue Waters, his team can now perform the same amount work in days instead of what used to take years. And for Blue Waters Professor Aleksei Aksimentiev, an associate professor of physics at the University of Illinois at Urbana-Champaign, the supercomputer’s scale is a requirement for accurate structure predictions and determining the molecular mechanism of a key step in the DNA repair process.
Christian Ott, professor of astrophysics at Caltech, said with access to Blue Waters, his team is now performing simulations that are pushing the frontier of supernova theory, enabling next-generation simulations with higher-resolution, including the first 3D global simulation of the MRI in rapidly rotating supernova. Brian O’Shea, assistant professor of physics and astronomy at Michigan State University, needs the system’s fast I/O and interconnects in combination with the large amount of memory to explore the earliest galaxies. It is computationally expensive work made possible through access to Blue Waters.
Leigh Orf, professor of atmospheric science at Central Michigan University, used 20,000 cores with over 100TB of data to simulate an EF5 tornado for the first time, which wouldn’t have been possible before Blue Waters. And Philip Maechling, information technology architect at Southern California Earthquake Center—which is studying the effects of earthquakes and assessing the maximum ground motions to help in establishing building codes—said that thanks to GPU optimization on Blue Waters, they were able to finish in 14 days work that previously took 61 days running on CPUs on both Blue Waters and the Stampede supercomputer at the Texas Advanced Computing Center at The University of Texas at Austin. This work also yielded predictions with much higher resolution than those from USGS.
Keynote Speakers Address Future of Supercomputing
William T.C. Kramer, director and principal investigator of the Blue Waters project, gave the opening keynote on the need for sustained petascale+ computing and data analysis. Recently, Kramer convened and led two community workshops (called Brainstorming HPCD) to identify the requirements from science and engineering communities for future high-performance computational and data analysis resources and services. Irene Qualters, director of the National Science Foundation’s Division of Advanced Cyberinfrastructure, also gave a plenary talk about where she sees supercomputing heading in the future and the paths it may take along the way.
Arden L. Bement, Jr, Davis A. Ross Distinguished Professor Emeritus and adjunct professor of the College of Technology at Purdue University with experience in executive positions in government, industry, and academia, gave a retrospective on the last three decades of NSF’s investments in academic supercomputing and the path already traveled.
The science community’s use of computation as a method of research has grown dramatically over that time period, a trend he believes will continue to evolve, “there will be more natural bridges between the human brain and the machine, helping to combine what the human can do best and what the machine can do best, faster.”
Bement said Blue Waters has shown this future has already started, “science and engineering are now able to solve more impactful problems: improving quality of life through better pharmaceuticals and drug discovery, through better diagnosis and understanding of chronic health problems and the connection with genetics.”
For Satoshi Matsuoka, professor at the Global Scientific Information and Computing Center of Tokyo Institute of Technology (GSIC), he believes demands for extreme computing and huge data processing is leading to a future with an inevitable convergence of the two infrastructures.
“Supercomputing is not just about providing the hardware resources; it’s about the software, the consulting, the libraries, the applications, and it is even about maintaining the open source. You get all that with Blue Waters, and the cloud lacks pretty much everything except the low-level hardware,” he explained. “That is why I believe you would find cloud vendors are eager to talk convergence.”
Scientists aren’t going to stop needing computation for their research, and they will seek it out in any form they can get it, big or small. In Matsuoka’s all-inclusive future, sustained growth in data capabilities, not compute, will advance the capacity and thus the overall capacities towards accelerating research and ultimately the industry.
Steve Scott, senior vice president and chief technology officer for Cray, wrapped up the symposium discussion of the future of big compute with his keynote on programming and technology for the next decade.
In the past few decades, simulation has become more and more important, and now every area of science has problems they can only address with simulation; Blue Waters gives them one of the most powerful tools in the world to do their work, as Scott emphasizes.
However, according to him, teams who are writing codes and using the Blue Waters system today need to start converting their codes to have multiple layers of parallelism. “Most people today are still writing straight MPI codes, and we need to have hierarchical parallelism,” said Scott. “And then the next step after that is paying more attention to how they exploit data locality and figuring out how to deal with deepening memory hierarchies.”
Scott discussed significant challenges—such as rapidly increasing on-node parallelism, varying forms of heterogeneity, deepening memory hierarchies, growing concerns around resiliency and silent data corruption, and worsening storage bottlenecks—that continue to surface for programmers as compute system continue to evolve.
“We need to start teaching programming as a parallel exercise from day one, so students learn parallel programming first, verses learning serial programming and then later trying to convert it.” He continued “and they need to be told what is expensive versus what is not expensive. Adding two numbers—not expensive. Moving data from one place to another—expensive.”
Presentations from the Symposium are available online at: https://bluewaters.ncsa.illinois.edu/symposium-2015-schedule-descriptions.