"New quality in HPC for science and research"
Rack from SuperMUC-NG: The programme "National High Performance Computing" adds the supercomputing landscape in Germany. Photo: V. Hohenegger/LRZ
More work for supercomputers at universities: powerful computer systems are no longer needed only for data-intensive modeling and simulations in the natural sciences or engineering. More and more new research areas at universities are relying on high-performance computing (HPC) to evaluate measurement and image data using artificial intelligence methods and to discover patterns in Big Data. Back in 2015, the German Council of Science and Humanities, an advisory body for the federal government and the states on issues in science, research and the higher education sector, suggested the "National High Performance Computing" (NHR) program. Selected computing centres at universities are to be able to expand their hardware capacities as well as consulting and educational services in the field of supercomputing in a more targeted manner. The NHR will now be financed by the federal and state governments with almost 63 million euros annually for ten years.
In the meantime, the NHR centres (see table) have been selected, they founded the NHR Association in 2021, which coordinates the tasks and funding and sets up a joint online platform for applications for computing time. The NHR further differentiates the German landscape of scientific HPC. The NHR association consists of nine university computing centers. Professor Dr. Gerhard Wellein, who heads the Centre for National High Performance Computing Erlangen (NHR@FAU) at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) and who's in the advisory council of Leibniz Suopercomputing Centre (LRZ) is one of three board members. In an interview with the LRZ, he describes the tasks and organization of the NHR - and how researchers apply for computing capacities.
How were the NHR centers selected? Wellein: Potential computing centres were pre-screened through a call for proposals by the Deutsche Forschungsgemeinschaft (DFG). Existing computing capacities and operating infrastructures played a role, as did expertise in HPC, training opportunities, excellency in user advice, and proven research foci in the application disciplines. The Gemeinsame Wissenschaftskonferenz (GWK), in which the federal and state governments align science funding, appointed a strategy committee with representatives from politics and science, which accompanied the NHR selection process and recommended the nine NHR centers. It will continue to monitor the NHR program. Incidentally, Professor Dr. Dieter Kranzlmüller from the LRZ is a member of the committee. Until May 2021 the Bavarian State Ministry of Science and the Arts was also a member. Bavaria should be a pillar of this new national HPC infrastructure – that was the wish of the researchers and the State Ministry from the very beginning. Bavaria therefore strongly supported the HPC activities in Erlangen and finally the application of FAU.
Most of the NHR centers also belong to the Gauß-Allianz, which also promotes supercomputing in Germany. Is it a partner or a competitor? Wellein: All NHR centres will of course continue to participate at Gauß-Allianz. However, the Gauß-Allianz is much broader in scope, with all three centers of the Gauss Centre for Supercomputing or GCS also participating, as well as other university and non-university institutions such as the Deutsche Klimarechenzentrum, the German Electron Synchrotron Desy or the Computing Centre of the Max Planck Institutes. We see the Gauß-Allianz as a partner for exchange between HPC centres in Germany and as a central information platform. The NHR can better be compared to the GCS. Both organizations or, more precisely, the centers organized in them carry the national HPC infrastructure that is open to researchers in Germany. The colleagues from the GCS cover the peak demand, while the NHR centres serve demanding projects in the breadth. In general, however, supercomputing in Germany is still much broader and is characterized by many local computing centres, institutes or research groups that cover the basic supply of computing power. This results in a great diversity, which the Gauß-Allianz represents in the best possible way.
How does NHR work with these organizations? Wellein: The NHR association was founded in August 2021. In the set-up phase, we are cooperating with both associations - on the one hand to avoid duplicate structures and on the other hand to ensure coherence and consistency in the use of HPC systems. For the management of application and allocation processes of HPC resources, the NHR will use the software JARDS, just like the GCS. The policy document for computing time allocation in the NHR was developed under the leadership of Professor Dietmar Kröner, who in turn has led the GCS process for years as chair of the steering committee. The boards of NHR and GCS will exchange information on a regular basis. The Gauß Allianz informs about events and teaching offers of NHR and GCS. NHR, GCS, and Gauß Allianz are also working closely together on plans to appear at the next ISC conference so that joint programs can be presented in close proximity. On a personal level, there is also much intertwining within the three organizations. Through the NHR, the HPC landscape in Germany will be further differentiated - closely intertwined with all national players.
The nine NHR-Centres
|
NHR-Zentrum |
Standort(e) |
Zugang zu Rechenzeit |
|
NHR4CES@RWTH |
Aachen |
|
|
NHR4CES@TUDa |
Darmstadt |
|
|
NHR@Göttingen |
Göttingen |
|
|
NHR@Zuse |
Berlin |
|
|
NHR@FAU |
Erlangen |
|
|
NHR@KIT |
Karlsruhe |
|
|
NHR@TUD |
Dresden |
|
|
PC2 |
Paderborn |
|
|
Konsortium Süd-West |
Frankfurt, Kaiserslautern, Mainz, Saarland |
For the NHR, the universities of Frankfurt, Mainz, Kaiserslautern and the Saarland have joined forces to form the South-West Consortium. Do you expect more such associations? Wellein: In fact, such concentrations of computing centres can currently be observed primarily at the state level and, in my view, these make a lot of sense. Science today needs computing capacity not only for data-intensive simulations, artificial intelligence and machine learning also demand high-performance computing. The same is true for expert consulting and training. Resources and know-how distributed throughout the state should therefore be coordinated and developed in a targeted manner to be available to all users in the region. NHR and GCS are then available nationwide for higher demands. The states of Baden-Württemberg and North Rhine-Westphalia have been moving in this direction for several years. Researchers and scientists thus have access to various technologies and broad expertise. To achieve this goal, Bavaria is taking a different approach. With the LRZ anchored in GCS and the FAU in NHR, Bavaria is closely integrated in both HPC alliances - with KONWIHR, we in Bavaria support researchers in being able to use computers and expertise at the national centers.
The NHR consortium has 62.5 million euros at its disposal each year - is that enough? Wellein: Financial planning is challenging in times when electricity price sky-rocket and supply chains are fragile. The NHR centres have some flexibility in spending items. However, higher energy costs and more expensive hardware now require a rebalancing of available capacity, services, and training offerings. A major goal of the NHR association is the exchange between the centers in order to adjust their planning in such a way that we can offer an optimal supply of hardware and methodological expertise throughout Germany within the budget. This is where the strength of such alliances becomes apparent. I am therefore confident that politics will continue to adhere to these goals and strengthen excellence in HPC in Germany. This contains a clear educational mandate. Also, given the need to develop scalable software for the next generation of exascale supercomputers now, we hope that funding programmes for researching and implementing HPC applications will continue to be launched. Collaborations with scientists and their project results are important for efficient use of the hardware of the future.
The NHR centres are intended to provide high-performance computing capacity to the scientific community: Can researchers apply for them yet? Where? Wellein: Similar to the GCS, researchers from all German universities have the opportunity to use the NHR. The NHR is currently setting up an online platform for submitting and managing requests for computing time. Until it is available, the nine NHR centers are accepting applications from all over Germany. To avoid duplicate structures, they have formed thematic and technical focal points and will therefore exchange applications as needed and after consultation so that scientists can use the systems that are best for them. The GCS focuses primarily on very large research projects requiring 25, 30 million core hours or more. The NHR collaborative caters to projects with less requirements. These are not sharp boundaries, but they provide guidance on where to find appropriate resources. Again, because NHR and GCS rely on comparable proposal processes, projects are referred to the appropriate collaborative when needed.
Physics, chemistry, and materials science are traditionally disciplines with high data volumes. In which other disciplines do you currently see growing demand? Wellein: The focus of the NHR@FAU applications is currently on atomistic simulations; they are needed in various disciplines such as physics, chemistry, biology, medicine and material sciences. In addition to the classic disciplines, we are currently seeing an increase in demand, especially in medicine and medical technology. This is certainly due to the fact that FAU has structurally promoted interdisciplinary cooperation between medicine, the natural sciences and engineering and has been able to establish the "Engineering of Advanced Materials" cluster of excellence. At FAU, but also at many other NHR centres, we see that the demand for computing capacity in the humanities and social sciences is increasing. This is primarily due to machine learning methods, which are bringing new disciplines and scientific disciplines to the computing centres in almost all NHR centres.
HPC is not yet part of everyday life in the humanities, social sciences, biology or medicine. How do you promote HPC in these disciplines that are more remote from computer technology? Wellein: New projects are still mostly the result of personal exchange, i.e. a chair or a research institute identifies a computer technology problem and seeks advice and a solution from the computer science department or the FAU computing centre. Often, the complexity of the applications then grows quickly, and the first HPC infrastructures come into play. The best example is a project from lingusitics that uses computers to research the connection between speech and gestures and has become one of the largest users of FAU's HPC resources in recent years. As at FAU, many universities are establishing collaborations with computing centers, especially in the digital humanities, to tackle new tasks. If the centers are involved in projects early on, they can develop specific workshops and training for researchers in these new HPC disciplines. One of NHR's goals, of course, is to encourage the involvement of new user groups.
What is your biggest wish for the NHR in the next 5 years, and what needs to be done now to make that happen? Wellein: If the NHR is significantly more than the sum of its parts by then, we will have accomplished a lot. The association should grow into a lively network that brings a new quality to science and research in HPC, hardware and software, and education and training. This requires the partners in the NHR association to concentrate on core competencies, to occasionally jump over their shadows and to delegate valued tasks to other centers because they are better advanced there. In this way, the NHR can develop into a nationwide association with distributed competencies - some may see this as a risk, but for me it is a great opportunity. (vs)
Prof. Dr. Gerhard Wellein, Director of NHR and Head of NHR@FAU