Technologies for high-performance computing continue to develop at a rapid rate, and the march toward faster, more powerful machines shows few signs of slowing down. Although such growth is good for science and technology, it also leads to increased physical needs for supercomputing centers — building spaces able to accommodate larger machines, cooling systems that efficiently manage greater heat production, and energy to keep everything running.
How can the need for more resources be managed in a sustainable way? In addition to improvements in supercomputing systems themselves, it is clear that the future must involve innovation in infrastructure for supercomputing centers, particularly infrastructure that can evolve as new computing and cooling technologies become available.
On October 23-24, 2018, IT specialists and research infrastructure experts from the German-speaking HPC community met at the High-Performance Computing Center Stuttgart (HLRS) to discuss holistic strategies for addressing such challenges. The event — the second annual meeting organized by HLRS on the topic of sustainability in supercomputing centers — revealed a wide range of ideas for sustainable computing architecture, while at the same time highlighting some important challenges and problems to be addressed.
Markus Raschka of the Baden-Württemberg Ministry of Finance framed the challenge faced by German universities in general, and academic computing centers in particular, in the effort to meet Germany's ambitious sustainable energy goals. Although the state has been making progress in improving its environmental footprint, Raschka reported, the extent of energy usage for computing in universities and other public offices suggests that additional gains must still be made.
Dr. Ulrike Kugler, who leads the Green-IT program at the Baden-Württemberg Ministry of the Environment, explained that the state has implemented a strategy for encouraging green IT infrastructure development. In her talk she outlined its goals as well as the consulting services and funding opportunities it offers for computing centers seeking to mitigate their environmental impact.
Barbara Schneider of the Staatliches Bauamt München 2, the state administration in Bavaria responsible for academic building construction, has worked closely with the Leibniz Supercomputing Centre, one of HLRS's partners in the Gauss Centre for Supercomputing. As she explained, the goal of any design for a supercomputing center is to have the newest technology installed when the center opens. But considering the fact that erecting a brick and mortar structure can take 5-7 years, the best technologies often only become available after a project is underway. This can result in changing requirements that can be difficult to integrate into a complex design and build process. Schneider emphasized that communication is an important foundation for any such effort, and advocated that all requirements for a computing center — including requirements that might arise later when new technologies or needs materialize — be articulated early in the design process. Finally, she advocated a new approach to funding supercomputer centers that would apply not only for the purchase of a computer, but also to the realization of an entire computing center concept.
Thomas Weyrich of HPC consultancy Akquinet underscored Schneider's remarks, noting that every computing center's project is unique and that no one set of guidelines can address all possible scenarios. He also pointed out that HPC centers are like the Formula 1 racecars of computing centers, meaning that they are always at the cutting edge of computing technologies. This presents special challenges to planning, however, because as he observed, "whenever you think you've reached a physical boundary, you discover something faster and more efficient."
An additional component of the workshop was a series of talks by representatives of high-performance computing centers in Germany, Austria, and Switzerland. The presenters focused on their experiences and accomplishments in trying to improve their computing centers' sustainability.
Among the techniques discussed during the presentations was the increased use of sensors for monitoring and optimizing supercomputing center operations. As Rolf Bogus of the University of Heidelberg reported, however, merging sensor data from the very different worlds of building and IT-systems still poses a substantial challenge.
Ernst Haunschmid of the University of Vienna described his center's use of immersion cooling technologies. Here, the computers are submerged in white mineral oil, which can efficiently absorb and disperse heat from computing nodes.
Tom Fieseler of the Jülich Supercomputing Centre explained that it is getting more and more difficult to house modern supercomputers that have grown in size and power capabilities in buildings erected several decades ago. Jülich recently introduced the concept of modular supercomputers, but as Fiesler explained, this is reaching the center's infrastructure limits and a new building will be necessary. One positive result will be that the use of modern technologies (direct liquid cooling, free cooling) will substantially increase cooling efficieny and thus reduce the center's cooling costs and environmental footprint.
Michael Ott described a four-pillar framework of the global optimization strategy at the Leibniz Supercomputing Center in Garching. He explained the idea of implementing an adsorption cooling system to reuse heat from cooling their supercomputer and detailed the pros and cons of this innovative concept.
Ladina Gilly described an innovative sustainability concept used at the Swiss National Computing Centre, which pumps water from nearby Lake Lugano to cool its computing system. As water falls through tunnels back to the lake, it also passes through turbines that generate power for the computing center and for nearby businesses. Additionally, the building was constructed with extra space that would enable the center to expand without the need to construct new facilities. Considering the complexities that the project had to address, Gilly suggested that sustainable building of supercomputing centers is only possible when IT professionals, architects and planners, local government agencies, and funders set out to address the subject from the beginning.
With all of these ideas for improving computing centers, how will the HPC community know that it is objectively becoming more sustainable? As HLRS's Brigitte-Maria Lorenz and Marcel Brodbeck explained, certification programs such as EMAS and the Blauer Engel offer a useful roadmap for systematically identifying steps that can be taken to approach these goals. In their presentation they discussed steps that HLRS is taking to achieve EMAS certification.
In the final talk of the workshop Frank Neubauer of CARMAO GmbH described a concept his team has been developing to address factors affecting sustainability throughout the lifespan of a supercomputing center — from manufacture of the system, through its distribution, use, and end of life. His group will be formally testing the framework in cooperation with computing centers in the near future.
— Christopher Williams