German HPC and EuroHPC, a question of competitive collaboration benefiting science

13 Aug 2018 Frankfurt - After his presentation at ISC'18 in the session "German HPC in Context",Primeur Magazinewas able to talk with Michael Resch, Chairman of the Gauss Centre for Supercomputing. His talk covered the German HPC developments, also in connection with the European developments in EuroHPC. Michael Resch toldPrimeur Magazinethat all these things fit nicely together because Germany has been very active. It was among the first countries to sign the agreement for the EuroHPC Declaration. This fits nicely, both to the German strategy and the German strategy fits nicely into EuroHPC. These are two very complementary activities.

Primeur Magazine:You also mentioned that in Germany the local structure of high-performance computing has been changed as well? Did I hear that correctly?

Michael Resch:No, we don't have to change the regional or the federal structure in HPC in Germany. It's rather that we have this layered concept. At the very low layer you have universities, locally responsible for HPC at a lower level. Then you have these regional or statewide centres which we call tier two. We work closely with these two levels - tier two and tier three - in Germany. Working together with them guarantees that researchers can actually move in this pyramid of performance from level 3 up to level 0, and maybe even down to level 3 when they do their big simulations and then, while they have to do the evaluation of their results, going back to their local university. The local structure, this federated structure that we have in Germany, very well supports the researchers. It actually allows us to have this hierarchical approach of the European, national, statewide, local level.

Primeur Magazine:There are three big supercomputing centres in Germany. During most of the history of the TOP500, these centres were somewhere in the top 10 or top 15. Now, they are a little bit behind. Is that their natural place?

Michael Resch:There is a gap now because we had a project in which we had the funding for systems from 2007 to 2019. We recently started a new project which gives us funding from 2017 until 2025, maybe 2026. The Jülich system came in slightly later than planned, so basically the big system in Jülich is probably delayed by one year. It is partially a delay in the funding and partially a delay in technological developments that we expected to happen. This is nothing that actually frightens us. The fastest system is ranked number 23. That is not good but we will be back in the game in November with the Munich system and in June 2019 with the Jülich system. Then you will see German systems in the top 10. This is a normal situation. Every big centre goes down and the good thing about us is that we usually don't go further down than probably rank 20. 2018 is very much an exceptional year.

Primeur Magazine:You also need exceptional years.

Michael Resch:Yes, sometimes you have exceptional years because you are better; sometimes you have exceptional years because it is like the worst but that's okay.

Primeur Magazine:In the plans for Germany, you also aim to have exascale systems around 2023.

Michael Resch:Absolutely. Professor Lucas from the Federal Ministry of Science announced this last year. We are working on a strategy where all the three centres move up to exascale level. We want one of the centres, namely Jülich, to be an exaflop centre for EuroHPC. This shows how we can merge the two strategies, the European strategy and the German strategy, Europe having the ability to fund two exaflop systems, Germany having the ability to fund much more than that. So, it will be a Jülich system with an exaflop and then Munich and Stuttgart will follow to reach this level probably slightly later. We will also be using different technologies to support different communities.

Primeur Magazine:Being part of EuroHPC means that from a German point of view, you could get much quicker access to exascale systems than you would have normally?

Michael Resch:I don't think it speeds up the process too much but there is more funding because you pull the resources in that case for Jülich. You pull the resources of Germany and Europe and by bringing the two of them together you reach the exaflop level much earlier than Germany alone could do. That is an advantage.

Primeur Magazine:In EuroHPC, they are talking about two things: first about making machines available for users and applications - and of course, that is a very important part which a lot of people tend to forget - and the other thing is to have exascale systems based on European technology. What do you think about these things: is one more important than the other?

Michael Resch:It is obvious that we have to support European science and European society. One approach is to provide the best technology. This means that you just look at the level of performance that scientists need, you look at the kind of support that they need for their simulations. That is sort of the number, the exaflop number. This is what you would like to reach to be competitive internationally when it comes to science and also when it comes to industrial usage. The other thing is this: Europe is a country or Europe is a continent - I call it a country, so I assume I am a good European - where we don't have big national IT players in the sense that the United States have an IBM or Japan has NEC. That puts us in a position where we have an advantage of competing companies but we have also a certain disadvantage of not having a local player.

Developing European technology for us as a centre means that we have a local player and there is more diversity in the market which is positive. These two things go together but it doesn't mean that if you have European technology we need to have exactly one exaflop or if we have exactly one exaflop it has to be only European technology. The target is to support science and society. This is an interplay of a variety of activities that are going on in HPC.

Primeur Magazine:The advantage of having European technology to some level is that also the industry can benefit of course.

Michael Resch:I don't know whether industrial people would see this the same way but I think they would be happy - as far as I work with industry - to have more choice. It is certainly easier for them to work with a European vendor in the same way as it is easier for us because of time zone differences and travel times if you have problems with things like that. However, I think that industry is less local than the European supercomputer centres are. Take an organisation or a company like Daimler. They have their development, research, production, manufacturing, and marketing whatsoever worldwide. They are not that much national anymore. This is something we see with big companies but also with small and medium-sized companies.

In the state of Baden-Württemberg we have companies with some 600 people working in 25 different countries all around the globe. Industry has a different view of this. For them it is not something like: it is either Europe or it is not Europe because they are anyway playing on several fields.

Primeur Magazine:Does this also hold for SMEs? You have a number of consulting companies, for instance?

Michael Resch:It depends. We work with medium-sized companies with 600, 800 or 1000 people. They have subsidiaries in some 25 countries. We also have small and medium sized companies that are only located around Stuttgart, for example. For them it makes a difference. The others are more international.

Primeur Magazine:You mentioned that Germany, and especially Jülich, would be a candidate to have one of the first EuroHPC machines. Barcelona would also be a candidate.

Michael Resch:Yes, this is good. It is a process of competition. Science is about competing ideas, science is about competition in a sense of having better technology, better ideas, better concepts, better strategies. I expect other candidates to come forward and hand in their candidacy. We will just see. The Gauss Center is in a strong position. We do excellent science, we have excellent systems, we have big centres, very good centres, so we are not afraid of a competition. In science, you have collaboration and competition. You could call it collaborative competition or competitive collaboration. That is normal.

I think this is one of the strengths of Europe. There is not a single government that probably makes a decision about this organisation being good and this organisation being not good. We are competing at the European level and we are competing, each of us, at the national level. That makes us better and that is positive.

Primeur Magazine:Thank you very much for this interview.

For more information visit the Gauss Centre website or the site with information on EuroHPC maintained by the Primeur Magazine team.

Ad Emmen