Looking Ahead to SC19
The 31st annual Supercomputing Conference (SC19) will begin on November 17 in Denver, Colorado. Dozens of attendees from LLNL’s Computing Directorate will join industry leaders, application developers, program managers, journalists, and congressional staffers for the six-day conference focused on high performance computing (HPC). This year, LLNL staff are participating in—and often chairing—more than 50 events including tutorials, workshops, panels, poster and paper sessions, and the Job Fair.
SC19’s theme is HPC is Now. See the SC19 Event Calendar for a day-by-day schedule of LLNL’s sessions and Job Fair information, as well as a list of our SC committee members. During the conference, follow LLNL Computing on Twitter and other social media with these hashtags: #LLNLatSC #SC19 #HPCisNow.
In addition to the technical program, SC hosts many events geared for students as well as exhibits that showcase the history of HPC technology. Notably, the latest Top500 supercomputer rankings will be unveiled at SC19. (LLNL’s flagship Sierra has held the #2 spot since SC18.)
Like many LLNL participants, computer scientist Maya Gokhale looks forward to multiple aspects of the conference. She says, “I most enjoy the panels, BOFs [Birds of a Feather], and networking opportunities. Meeting students and academic researchers to discuss research ideas in depth is a big plus. Mentoring at SC is important to me, and I try to participate in mentoring sessions.”
John Gyllenhaal, who is developing LLNL’s HPC user environments for the Sierra supercomputer and upcoming El Capitan system, also emphasizes the networking value of SC. He states, “The conference provides face-to-face opportunities to discuss technical content. When you meet a vendor or potential collaborator, you have context for each other when interacting later on the phone or in other ways.”
Figure 1. The Department of Energy’s booth at SC18 highlighted HPC accomplishments from the labs and featured live demonstrations and talks.
Tutorials and Workshops Aplenty
LLNL researchers lead 14 SC19 tutorials and workshops, several of which have been offered in previous years. For example, the Computational Approaches for Cancer Workshop returns for the fifth time, while the Energy Efficient HPC Working Group opens its tenth workshop on SC’s second day. Other tutorials and workshops plumb the depths of OpenMP, in-situ analysis and visualization, HPC performance, and HPC software correctness.
Gokhale chairs and moderates the Workshop on Memory Centric High Performance Computing—known as MCHPC’19—on November 18. As co-organizer and session chair, she worked with the organizing committee to define a workshop theme and to solicit papers and invited speakers. As panel moderator, she will introduce the speakers, keep the panel on schedule, and field audience questions.
During MCHPC’19, participants will explore challenges associated with the increasing complexity of HPC memory systems. “We would like to draw practitioners and researchers together to discuss challenges, approaches, and opportunities in navigating the complex memory hierarchy,” states Gokhale. “With the intertwining of data science and physics-based simulation, advances in memory technology offer unique benefits. However, we are still learning how best to use these new resources. We hope the workshop will help fill that gap.”
The ever-popular Spack tutorial returns on November 18. The tutorial had 60 attendees at SC18 and 75 at this summer’s Practice and Experience in Advanced Research Computing (PEARC) Conference. Peter Scheibel is one of six presenters who will usher participants through installing and configuring the package manager, using Spack to create environments, and more. He explains, “We break the tutorial into two parts—beginner and advanced concepts. Folks don’t actually have to attend the full day, but if they do, we provide a gradual increase in complexity.”
The large, international Spack community will also assemble at a BOF session a few days after the tutorial. Scheibel says, “I’m looking forward to the BOF because it helps gauge the popularity of specific features. Sometimes people have interesting workarounds to problems that I assumed would take a great deal of coding effort to solve, and they generally mention it in response to a question from someone they’ve never met. It’s hard to imagine how that information could be readily shared without gathering the Spack community into the same room.”
Figure 2. Cyrus Harrison introduces an in-situ visualization tool at an SC18 workshop.
Best Paper Finalist
Of the several dozen technical papers accepted at the conference, “A Massively Parallel Infrastructure for Adaptive Multiscale Simulations: Modeling RAS Initiation Pathway for Cancer” is one of nine finalists for Best Paper. Computing’s co-authors are Frank Di Natale, Harsh Bhatia, Peer-Timo Bremer, Tom Scogland, and Fred Streitz. Nine Physical and Life Sciences researchers and Computing alumni Shiv Sundaram and Liam Stanton also contributed, as well as researchers from Frederick National Laboratory for Cancer Research, Los Alamos National Laboratory, IBM, and San Jose State University.
The team has developed a novel, massively parallel multiscale simulation infrastructure that uses machine learning to combine a macro time/length-scale model with a microscale (molecular dynamics [MD]) model to provide detailed insights into complex biological phenomena on heterogeneous HPC architectures. The paper demonstrates the scaling of the infrastructure to the whole of Sierra, with almost 16,000 concurrent MD simulations, aggregating a total of 120,000 simulations over several days, resulting in almost 200 milliseconds of MD trajectories—orders of magnitude ahead of the state of the art. In other words, Bhatia says, “We ran a simulation at an unprecedented scale, which required effort at an unprecedented scale from the team.” (Read the abstract or full paper.)
SC is a highly prestigious venue for technical research. “Just having a paper accepted at SC is a big deal,” explains Bhatia, who will be attending for the second time. “Special recognition, such as in the Best Paper category, is a great opportunity for everyone—great publicity for the cool work we do here at the Lab, as well as for the individuals who contributed to and made the success possible.”
Twenty-seven LLNL employees serve on SC19 committees ranging from support and infrastructure teams to technical program tracks. Several staff are reprising roles held at SC18 and previous events, while others are new to committee representation.
In one new appointment, Lindsey Whitehurst was named to the SC Network Security Committee, which is responsible for protecting attendees, visitors, and exhibitors by mitigating network security threats. This includes monitoring and filtering network traffic, as well as detecting and preventing compromised systems. Bronis de Supinski serves both on the SC Steering Committee and as Finance Chair for the Executive Committee. In 2021, he will be the overall Conference Chair.
Advice from the Pros
Many LLNL staff are veteran SC attendees who willingly pay their knowledge forward to first-time participants. Gokhale, who has been to at least 15 years of SC, recommends planning ahead: “With so many opportunities, it’s easy to overlook an event you really want to see.” She also advises meeting and talking with a range of people, noting, “Chance meetings and conversions can lead to new ideas.”
Heading into his 19th SC, Gyllenhaal agrees, “Be aggressively outgoing.” His penchant for wearing large, whimsical hats to the conference is one way to break the ice. (New SC participants can look for him in a giant foam hat near the Department of Energy exhibit booth #925.)
Furthermore, the size of the event can be overwhelming. Scheibel points out, “SC has nearly 15,000 people, and the venue is correspondingly large. It can take a significant amount of time to go between events.” Gyllenhaal adds, “Try to get enough sleep.”
Figure 3. Tony Baylis, John Gyllenhaal, and Jean Shuler pose in Texas-sized hats that Gyllenhaal brought to SC18, held in Dallas.