In this interview, Alex Heinecke from Intel and Sebastian Rettenberger from the Technical University of Munich speak with Rich Brueckner at ISC’14 about winning the 2014 PRACE award.
insideHPC: Tell us about the work for which you were recognized – what is the science all about?Alex Heinecke: It was basically the eruption of a volcano. It was executed on one of the biggest machines in the world– the SuperMUC, which is positioned 12th on the TOP500 list. We used all 150,000 cores that are available and we got a sustained performance of more than one petaflop.
insideHPC: More than one petaflop? What were you simulating about the volcano?
Alex Heinecke: The eruption basically, and how the waves are traveling through the mountain when you have this eruption.
insideHPC: Does this help you learn about safety and other aspects of volcanoes? Is that the objective?
Alex Heinecke: Exactly. It helps us to understand how we can have such seismic waves traveling through the soil. And then, in the long term, do real earthquake simulation in order to protect our assets here.
insideHPC: Sebastian, what was your role in this?
Sebastian Rettenberger: I was working more or less on the input, as we have a large mesh. You need to read this mesh into all of the code of the cores and to memory, and it’s not easy to read a large file on hundreds of thousands of cores.
insideHPC: Sebastian, I want to ask you about the code. You’re associated with Intel’d Parallel Computing Centers program. This is all about code modernization. So was there a baseline code that you had to port to the many core architecture?
Sebastian Rettenberger: Yes, basically. The actual code is 10 or 15 years old already. The Informatics Department is now trying to modernize it, vectorize it, and use all the advanced functions of a modern CPU to improve this application and make it run faster.
insideHPC: I have to say, when I thought about PRACE Award Winners I was expecting a couple of old guys. You guys seem very young to be doing groundbreaking science, is there a lot of that going on at your organization?
Alex Heinecke: Yes, we have a focus on high performance computing in our group. For me it’s the second PRACE Award in a row because we won the 2013 Award on Molecular Dynamics. So we basically say, “You need to optimize the entire simulation pipeline.” We’re also working together with “older” guys from the Geoscience Department or from the Supercomputing Center and we are the PhD students responsible for doing the work here.
insideHPC: I want to ask you about the challenges of extreme scale. Did you develop it on a small system and then work with it to get it to scale on this gigantic machine – this SuperMUC?
Alex Heinecke: Yes. We identified some bottlenecks and then determined how to remove these bottlenecks. As Sebastian has mentioned, it’s not just about optimizing the compute part of the application. You have to care about I/O and you have to care about initialization because the Computing Center wants to ensure you utilize the machine for computing and not for reading files.
insideHPC: What’s the next step with this? Is it a bigger problem or more fine mesh? What would be the next logical step from what you’ve achieved here?
Sebastian Rettenberger: The next step is the bigger mesh. And then also, thinking about how we can leverage accelerator technologies. Because as you see we were computing for three hours for this simulation and it was just a volcano — not a real earthquake. A real earthquake has even higher computing demands. We need to think about using core processors together with the main processor to accelerate the simulation and then solve a large scale earthquake science problem.
Watch the video with Rich, Alex Heinecke from Intel and Sebastian Rettenberger from the Technical University of Munich.