“Get the best out of the data” with high-performance computing

Infrastructure alone isn’t enough, says the head of high-performance computing at the University of Luxembourg, Pascal Bouvry Photo: Guy Wolff/Maison Moderne

Infrastructure alone isn’t enough, says the head of high-performance computing at the University of Luxembourg, Pascal Bouvry Photo: Guy Wolff/Maison Moderne

Professor Pascal Bouvry, head of high-performance computing (HPC) infrastructure at the University of Luxembourg, looks at the master programme coming in 2022, developed with a team of European partners.

The HPC master’s degree is a European project. How did Luxembourg come to lead the initiative?

Pascal Bouvry: You hear about HPC everywhere, about HPDA [high-performance data analytics, editor’s note] and AI. The key thing now is to define the body of knowledge necessary to master this field. We’ve been working with a large set of partners to start defining this, what’s going to be included, build the lectures. It’s really a prototype. We have quite a few prestigious partners in this consortium, including the Polytechnic University of Catalonia, the Polytechnic University of Milan and the Sorbonne. We are going to define specialisations that correspond to niche markets. Luxembourg has been extremely proactive in the field of HPC. We’re also a very small country, meaning that we can put the actors easily around the table and make fast decisions. Also, with a young university, we can be more agile.

What kinds of skills will graduates bring to the job market?

As I mentioned, there will be different specialisations and also a match with other markets. For instance, some countries work more on the hardware side. Luxembourg is more about the data economy. We just started a new master’s in data science, we’ve been running a master’s in computer science. We can see the master’s in HPC as putting these two together, providing software engineers who can develop efficient code for the HPC, but also data scientists that can take advantage of this kind of machine.

When you say “advantage”, how is HPC shaping research and innovation?

HPC is really helping to provide a new dimension. Imagine the kind of granularity you can achieve, for example for designing new materials. With HPC, you can go for something that is more robust, well-defined and later on requires less testing in real life, cutting costs by providing high-level simulation and modelling. We’ve been gathering a lot of data. Space is a booming market. At the Luxembourg Space Agency there are already something over 20 petabytes of data. By using Meluxina, we can really get the best out of the data.

To what extent is this kind of technology essential for Luxembourg to be competitive?

It’s not only essential to be competitive but also essential to have sovereignty over the data and data processing. For some of the data and some of the processing, you don’t want this going across borders, you don’t want people to be able to peek inside. It’s essential that Luxembourg has the right set of infrastructure, but infrastructure is not an end in itself. We need the expertise and the right set of people to operate the infrastructure and bring solutions to our industry and society.

The US is looking to launch its first exascale computers at the end of 2021 and in early 2022. What does that mean for the technology we have in Luxembourg right now?

When I was doing my PhD, the target was a gigaflop. Now, we talk about peta or exa. The supercomputers of yesterday are today’s desktops and will be tomorrow’s mobile phones. We are working in a tier architecture, depending on the needs. You need to find the right platform. In some years, the exaflop will be hitting Europe. The newer set of funding for EuroHPC--something like €7bn--includes making calls for exaflop machines. The way we see current needs, Meluxina is a very good fit but in a couple of years, new demands will come. In 2006, I worked with biologists and they were happy to sequence a family--two parents and two children--and today they are sequencing populations.

This article first appeared in the January 2022 special forecast edition of Delano magazine.