"Do Good and Teach about It": HPC Training at HLRS

The approximately 1,000 square-meter, €6.8 million facility offers a state-of-the-art lecture hall, smaller seminar rooms, and space to expand HLRS staff. It is already having a major impact on the quality of the training activities, conferences, and symposia that HLRS can offer.

Speaking at the inaugural celebration, Baden-Württemberg State Secretary for Finance Gisela Splett called HLRS a "lighthouse" for advanced science and research and hailed the opening of the new HLRS training facility as an event that will enhance the center’s global reach. "This will be a meeting point for a network that will play an important role in future generations for the scientific and industrial competitiveness of Baden-Württemberg and Europe," she said.

The opening of the new HLRS training center has not been the only improvement in the center’s training activities in recent years, however. By cultivating collaborations within its extensive network and integrating new pedagogical approaches and education technologies, HLRS is pursuing a multifaceted strategy aimed at disseminating the technical knowledge necessary to use high-performance computing.

"You all know the saying 'Do good and talk about it,'" said Simone Rehm, Chief Information Officer at the University of Stuttgart, at the inauguration ceremony. "I’d like to modify the saying and suggest we should 'Do good and teach about it.'" Indeed, by making training a core component of its mission, HLRS is amplifying the impact of simulation technologies in research and development.

Numerical simulation has long provided powerful tools for R&D. However, running simulations on supercomputers requires specialized knowledge that many scientists and engineers do not gain during their academic training. According to Rolf Rabenseifner, who oversees the HLRS training program, "Our goal is to address this niche, providing continuing professional education that delivers the essential practical skills that researchers need when using simulation to solve complex problems."

Because HLRS is a German national supercomputing center located at the University of Stuttgart, the overwhelming majority of its trainees are academic scientists working on publicly funded research. When young investigators join a lab that works on problems in aerodynamics, climate modeling, or molecular dynamics, for example, the lab will often send them to HLRS to gain the HPC skills necessary for that lab’s research.

For this reason—for more than 20 years—the core HLRS training program has focused on the most important basic skills that HPC users need: parallel programming interfaces such as MPI and OpenMP, as well as Fortran and advanced C++ for high-performance computing and HPC architectures. Over the years its portfolio has grown to include other related topics such as cluster file systems and performance optimization, as well as specialized HPC domains such as visualization and computational fluid dynamics.

The curriculum that HLRS offers continually adapts to the latest HPC trends, tools, and challenges. In 2017, for example, it hosted a two-day workshop organized by Cray (the builder of HLRS’s supercomputer, Hazel Hen) and hardware manufacturer NVIDIA that was designed to demystify the up-and-coming fields of artificial intelligence and deep learning. In the Stuttgart region—home to a large high-tech community surrounding auto manufacturers Daimler and Porsche—these fields have attracted interest because of their relevance for future self-driving vehicles. The course was the most highly attended of the year, covering topics including object detection, image segmentation, and neural networks.

“Supercomputing and its applications are constantly evolving and we need to stay current with these trends,” says Rabenseifner. “In the near future, for example, we anticipate a convergence between data analytics and high-performance computing that will present new kinds of challenges. Staff at HLRS will be on the front lines of navigating this convergence and our training program will enable us to share the knowledge we gain with people in both academia and industry whose work will benefit from it.”

HLRS's training activities are not aimed only at local researchers, but constitute an ambitious effort to build HPC know-how in Germany, across Europe, and internationally.

As one of three members of the Gauss Centre for Supercomputing (GCS)—an alliance that forms the backbone of Germany's national HPC infrastructure—HLRS coordinates HPC training with GCS partners the Jülich Supercomputing Centre and the Leibniz Supercomputing Centre, as well as other German supercomputing centers at the Technical University Dresden, University of Siegen, and University of Frankfurt. Some training courses are held in Stuttgart, while others are hosted offsite to make HLRS knowledge more easily accessible to investigators in other cities.

In addition, as a member of the Partnership for Advanced Computing in Europe (PRACE), HLRS is a PRACE Advanced Training Center, providing state-of-the-art courses for scientists across the continent. In addition to offering traditional classroom learning HLRS is also helping to develop a massive open online course (MOOC) covering parallel computing.

To increase the reach of technical expertise hosted at HLRS, the center also offers a "train the trainer" program focusing on parallel computing. Designed with HPC instructors located at other institutions in mind, the program not only covers the latest standards for MPI and OpenMP, but also provides trainers with pedagogical strategies for conveying this knowledge more effectively to their own students. Thus far, HLRS's train the trainer program has enabled the development of new courses at supercomputing centers in Belgium and Austria, and in 2018 plans to expand this effort in the Netherlands and Ireland.

HLRS’s international training collaborations also include a biannual, 10-day German-Russian Young Scientists' School and Conference on Parallel Programming and High-Performance Computing, held in Novosibirsk, as well as tutorials at major annual supercomputing conferences such as ISC High Performance in Frankfurt and the International Conference for High Performance Computing Networking, Storage, and Analysis (SC) in the United States.

Maintaining this network has an important online component, as HLRS distributes updated course materials through the center's website. "By making electronic recordings and slide sets of previous courses available," Rabenseifner explains, "we give researchers anywhere, regardless of geography, access to useful training information when they need it." This "just-in-time" delivery approach enables scientists to gain knowledge when they need it and allows course participants to review material they might have been exposed to but didn't completely master in the classroom. Maximizing flexibility in this way helps scientists to navigate the significant challenges that come with HPC.

Although the traditional HLRS trainee works in academic research, simulation is also extremely important for the ecosystem of small and mid-sized precision engineering companies that make Baden-Württemberg an economic powerhouse. Many of these companies, however, are not large enough to support full-time, in-house HPC experts able to keep up with best practices in programming and data management.

In 2016 HLRS set out to address this critical knowledge gap. With grants totalling €2.5 million from the European Social Fund, Baden-Württemberg Ministry of Science, Research and the Arts, and the University of Stuttgart, HLRS launched a project to develop a specialized training program targeting industry’s unique needs. The initiative is being developed in collaboration with partners at the University of Freiburg, the University of Ulm, and Sicos BW, and opened to trainees in 2018 as the Supercomputing-Academie.

“The Supercomputing-Academie is the next generation of continuing education at HLRS,” says Andreas Wierse, CEO of SICOS BW, an independent organization based at HLRS that facilitates access to HPC for small and mid-sized companies. “Just as the new training building at HLRS is physically connected to the building that houses Hazel Hen, the Supercomputing-Academie will be an integral, core part of HLRS’s training operations.”

During curriculum planning for the Supercomputing-Academie, HLRS set out to address several limitations that have made it difficult for industrial HPC users to take advantage of its existing training program. “From the beginning,” says Sicos BW's Markus Klietmann, “we engaged in a dialog with industry representatives to find out what their needs are and how the expertise that exists at HLRS could best help them.”

One important practical consideration was that industrial HPC users typically don’t have the time to dedicate days or weeks away from their job to spend in a traditional classroom. For this reason, the Supercomputing-Academie is using an educational approach called blended learning, which combines the best qualities of classroom learning and online learning. Trainees meet the instructor and students onsite at HLRS, though the majority of instruction happens online. In addition to independent study that participants can do at their convenience, there are virtual classrooms where they can meet over the Internet each week, exchanging ideas and asking questions of the instructor.

Considering that the typical industrial user works 8 to 10 hours a day, has family commitments, and then needs to sit down and learn HPC, the flexible approach that blended learning offers is meant to make this as easy as possible. The interactive dimension also gives participants valuable cross-industry perspectives that they wouldn't gain otherwise.

The Supercomputing-Academie launched in April 2018 with a module focused on parallel programming, and the HLRS team is in the early stages of developing additional modules that will appear in the coming years. Future module topics are likely to include HPC cloud computing, performance optimization, economic and ecological aspects of HPC, simulation, visualization, and data management.

An additional interesting dimension of the Supercomputing-Academie’s curriculum is its modular nature. This was also done intentionally based on interviews with industry representatives that revealed several different kinds of HPC users—engineers for whom HPC is an important component of product development, programmers whose code must be optimized to run on HPC systems, and computing resource managers who must ensure that staff at their company has the computing resources it needs. Each type of user has different interests and so the Supercomputing-Academie aims to be like a chameleon, changing appearance based on an individual trainee's needs.

"Knowledge and experience with HPC gives a company certain advantages," says Hanna Skubski, project coordinator for the Supercomputing-Academie at HLRS. "The Supercomputing-Academie will transfer the knowledge of HLRS experts to industry in ways that will ultimately benefit their ability to innovate."

Data analytics, simulation, and visualization will grow in importance for basic research, precision engineering applications, and other fields in the coming years. Taking advantage of these tools will mean that HLRS needs not only to maintain a first-class supercomputing infrastructure but also to empower people to use it efficiently and effectively.

“The new HLRS training center has been a major step forward for us,” says HLRS director Michael Resch. “But at the end of the day, a building and a supercomputer are only as valuable as what people do with them. In this sense, continuing to develop our training offerings—especially for industry—has to be an absolutely essential part of HLRS’s activities. They will continue to ensure HLRS’s place as a leading scientific center for simulation technology in Germany and in Europe, and will ultimately enable the scientists and businesses that work and train here to be as globally competitive as possible.”

—Christopher Williams