Award - PhD thesis
Matthieu Dorier wins the 2015 Gilles Kahn Thesis Award for his research with the Kerdata team
Matthieu Dorier, accessit du prix de thèse Gilles Kahn 2015
Matthieu Dorier has received one of the two 2015 Gilles Kahn Thesis Awards - presented by the French computer science society SiF, under the patronage of the French Academy of Science - for his doctoral research with the Kerdata project team at the INRIA Rennes Bretagne Atlantique research centre. Interview.
Matthieu, can you tell us about your academic career?
I was admitted to École Normale Supérieure in Cachan/Rennes in 2008. After an initial internship with the PARIS team at the INRIA facility in Rennes in 2009, reporting to Luc Bougé, I moved to the the University of Illinois in Urbana-Champaign (UIUC) in 2010, where I researched data storage for high-performance simulations. This internship, supervised by Marc Snir and Franck Cappello, took place at the Joint INRIA/UIUC Laboratory for Petascale Computing, where I had the opportunity to work with international partners in ideal conditions. This facility has since grown into the Joint Lab for Extreme-Scale Computing (JLESC), which also includes the Argonne National Laboratory, Barcelona Supercomputing Center, Jülich Supercomputing Center and Riken AICS. I ended the second year of my Master’s degree in Rennes with an internship with theKerData team, continuing my research as part of my doctoral studies under the supervision of Gabriel Antoniu and Luc Bougé. Having visited Argonne National Laboratory several times in the course of my thesis and completing an internship there, I chose it for my post-doctoral studies, reporting to Rob Ross.
You recently won one of the two Gilles Kahn Thesis Awards from the Académie des Sciences in recognition of your thesis with the KerData team. Could you tell us about your research and your three years working on your doctorate?
My research focused on data management for digital modelling in high-performance computing applications. Digital simulation has become an essential tool in modern science and industry; simulations are used to model the climate, viruses, nuclear fusion processes and even the birth and evolution of the universe. In industry, digital models are increasing replacing real prototypes, as they can be produced more quickly and cost-effectively, and are reproducible. Such simulations require considerable computing power, as delivered by the millions of processor cores in a supercomputer. Unfortunately, such simulations also generate vast quantities of data that are a challenge to subsequently store, analyse and visualise. However, this “data storage and processing” task is an essential prerequisite for understanding the modelled phenomenon and gaining scientific knowledge. My work during the thesis addressed this challenge by developing methods to enable the “data storage and processing” task to be performed more quickly. I began by devising solutions to “mask” the impact of data accesses on simulation performance, facilitating simulation scaling. I also created methods for visualisng simulation data by retrieving it directly from the supercomputer’s memory while the simulation is being executed. This paradigm is referred to as “in situ visualisation”. All these contributions were implemented in a software application named Damaris, which I was fortunate enough to be able to test on some of the world’s most powerful supercomputers, including Kraken, Blue Waters and Titan. I went on to research the congestion caused when concurrent applications access data: as supercomputers are generally used by multiple users simultaneously, it is important to understand how and why an application’s performance may be impacted by another application running in parallel. Lastly, I did some research in the area of simulation behaviour modelling, with the aim of predicting data accesses. As I said, all this research was conducted within the framework of joint initiatives, in particular between INRIA, UIUC and ANL, overseen by JLESC. During these three years of postgraduate study, I have built a sizeable network of colleagues with whom I still work today. This award is not just a source of great pride for myself, but is also a form of recognition for my coworkers and the research we are doing together.
Where do you see your career heading from here?
I am still passionately interested in academic research, and am currently doing postdoctoral research at Argonne National Laboratory. ANL is a great working environment, where I can develop new skills that at some stage I would like to share with the European research community, especially in France!