Faces of HPC: Frauke Gräter

Frauke Gräter was interviewed by Tiffany Trader, HPCWire on 18 May 2017.

What is the work of the molecular biomechanics group at Heidelberg University?

We work in one part on materials that nature has made that are fascinating and also from an application point of view interesting, which is silk and nacre – actually this award was partly because of our work with nacre. Nacre is also known as mother of pearl. It’s mainly just like chalk, so a very cheap material, calcium carbonate that you find everywhere. It’s very regular and on a nanometer scale. There are protein layers in between these crystal tablets and this makes nacre so special. With the computer we model the interactions of the living net of protein of the inorganic material, the calcium carbonate. This shiny material inside shells is mechanically so robust – it’s a way that shells protect themselves.

What are some of the real-world benefits of your research to society?

Industry tries to substitute other materials like steel with more lightweight materials, also bio-compatible materials. Both nacre and silicon are candidates, not necessarily to use them directly but if you understand the way nature has built them, to mimic these features, these strategies in synthetic materials, that’s the attempt — and computer simulations can very much help in seeing what are the screws we can use to make artificial materials better than they are now.

We are seeing more attention to diversity and inclusion in HPC — what can the community do to encourage broader engagement in science and computing (STEM/MINT*) fields?

I think efforts on all levels are needed, and I do see them happening. We just had a girls day at the institute where school girls would come and look at how to work at the computer because I think in many cases that boys really like to play more computer games than girls. That’s one entrance way. Then they start to program and think it’s fun and want to do science with it. You don’t find this for girls as much; you find them mostly interested in the topics. They want to learn about how this protein in your body works and then the computer is a good tool one can use to explore. Girls are in medicine and biology, they want to know the mechanism in the body, so in the lab you find women and then they come to computers as a very good tool to learn even more. This is a way that I think we can attract females into the field, by the scientific question not so much by pushing them into the next programming class, so to say. First the motivation, then the programming – more by curiosity, this is what was my motivation.

Can you point to positive signs and actions you are witnessing?

So in Germany, these girls days have been implemented in many cities and universities, and then there are some mentoring networks available, especially to women early career researchers, so I see many initiatives. It still really hasn’t gone all the way, of course, but it needs time. I do see positive developments. I for example always had a fairly okay male-to-female ratio in the group; and I think being a role model in the lectures myself is helpful.

Are you seeing greater involvement of women at events, on panels?

I think it is now expected to have a woman on the panel – it’s a good thing, but I think a quota here would be very problematic because you might end up with a woman that’s not in terms of her expertise considered appropriate for the panel. I think this consciousness is there so it happens, but I’m not fond of quotas for these things. It can make things go backwards; this I see problematic.

You participated in the closing panel [at PRACEdays], “The gap between scientific code development and exascale technology” — would you share some of your thoughts on that topic and on how your research codes will benefit from exascale?

As was brought up, the cases where you have a scientific question that really needs exascale are rare so for my field actually there is now the attempt to simulate whole cells and at the moment we still do single protein. [There’s an opportunity] to make it more complex and big when we have exascale: let’s simulate the cell, but at the same time, the time-scales of interest go from nano-seconds and micro-seconds to needing to simulate for seconds, and so solving one problem can become much more complex. You all of the sudden need to be more accurate to extrapolate to even longer time scales, so this will be very hard to come up with well-posed scientific problems that you can validate and so forth that make use of an exascale computer. I think we will fill these exascale computers up and we will learn through that, but it will be a learning curve by actually doing these calculations there.

So if you have a petaflops machine, what percentage of that machine do you need at this point?

We often actually can easily use even a small fraction, 20 percent; if we have a long time slot, many CPU hours, a fraction at a time is totally fine – so we need millions and millions of CPU hours easily but the scaling must not be such that we fill up the whole computer – so this embarrassingly-parallel way is a way we can easily do happily for our scientific question, but that’s not exactly what an exascale machine is for. So here I see a vulnerable point.

I think there is a chance from my field there will come applications, but it needs very careful thinking about which ones.

Coming off that question, what are the other gaps or pain points you see in HPC? What developments or technologies would most further your research?

I think machine learning will move into my field and this whole question of data storage is an important one on the technical side. Not the I/O during the simulations, but the data handling in post-processing because that’s also what you first of all don’t apply compute time for, so you’re left with your data and the analysis is not yet developed such that you would do this in parallel and then it’s distributed and you want to visualize it – so a cloud type of high performance computer with a very high speed interconnect is something we will also need at that point.

In the closing panel, you spoke of AI as an exciting development that is also important for attracting young people into the field. How will you bring AI into your workflow?

I see the first people just doing it – you ask colleagues and they are doing AI. In our case, there is opportunity for improvement of parameters by AI, also the analysis of data and substituting simulation by AI, so I see various aspects; it will never make our physics based models disappear but it’s a complementary way for us to advance.

*STEM/MINT: STEM is a popular but not universal acronym referencing the fields of science, technology, engineering and math. The equivalent concept in Germany is “MINT,” which stands for math, information technology, natural science and technology

Read the full interview with Frauke Gräter at HPCWire.