Marta García-Gasulla is an activity leader at the Computer Science department of the Barcelona Supercomputing Center (BSC), which she joined in 2006. She obtained her PhD in Computer Architecture from the Universitat Politecnica de Catalunya (UPC) in 2017. Her research interests include load balancing, parallel programming models, performance analysis, and optimization.
She co-leads the BSC’s Best Practices for Performance and Productivity (BePPP) activity group. With BePPP, she aims to bridge the gap between scientific domain researchers and computer scientists researchers. Promoting best practices for programmers to productively (re)structure their codes in ways that can result in high efficiency and portability. She also captures the fundamental co-design input to the appropriate system software or architecture team to target their developments in the most helpful direction. This goal is materialized through collaborations with standardizing bodies, like the OpenMP language committee or European flagship projects like the European Processor Initiative and several European HPC Centers of Excellence. In this third interview of the Meet the Researcher series, learn about Marta’s role in the ChEESE2 Project and what inspires her most about her work in HPC.
What are the main tasks of the BSC in the project?
BSC has several groups from different departments involved in the project. In particular, the teams involved from the Computer Science department, where I belong, are in charge of two differentiated tasks. On the one hand, we provide a continuous performance monitoring tool called TALP (mole in Catalan) to be used by the different codes. TALP is being further developed within the project to better support the needs of the flagship codes of ChEESE2.
On the other hand, we are also in charge of the co-design task with the EPI accelerator (European Processor Initiative). In this task, we test and analyze the project codes in the available platforms or prototypes of the European accelerator; we focus mainly on the vectorization capabilities. From this analysis, we provide feedback to the code owners on how to restructure their code to achieve better performance in these platforms, and we can also gather insight for the software and hardware developers of the EPI project.
Reflecting on the past year, what moment stands out as particularly significant for you and the project?
Being new at ChEESE2, the most relevant moment was the kick-off meeting, where I could meet all the partners, present our work plan, and especially learn about the HPC challenges we want to address with the project.
Can you explain how performance analysis of parallel HPC applications play a role on hazard assessment?
I always say performance analysis is crucial nowadays; the more powerful the supercomputers, the more important it is to make efficient use of the computing resources to achieve faster simulations and save energy and power. However, in the case of hazard assessment, it is even more critical, as the time to solution is much more important in this kind of simulation because it can mean predicting a hazard on time or not.
Are there any particularly challenging aspects or obstacles that you’ve encountered in your research in ChEESE, and how have you overcome them?
One of the issues we have found is that we needed to do a first performance analysis of the codes during the first year of the project, but the codes were not deployed in the EuroHPC systems on time, partly because the access to the systems was delayed. Having a large number of codes in the project makes it even more critical because we had to analyze all of them. We managed to deliver by decreasing the detail of the analysis for the deliverable, doing the detailed analysis later, and working coordinated with several partners to achieve the goal.
How do you foresee the impact of your research on European society?
I aim to help developers be more aware of the performance of their codes and make that performance portable.
As a professional that works with HPC, what inspired you to pursue this specific area of research, and what excites you most about your work?
During the last stages of my PhD, while I was testing my developments using a CFD (Computational Fluid Dynamics) code, I realized that we, the computer scientists researching new techniques to advance HPC, were not taking into account what the users of HPC need, and on the other hand, the code developers were trying to solve those problems on their own. That is how I started to, through the performance analysis of the codes, try to bring these two communities closer. The most exciting part is working together with scientists from different scientific domains; I would love to learn about their simulations and use cases and be able to help them.
What would you be if you were not a researcher?
It isn’t easy to imagine myself doing something different, but I guess if I were not a researcher, I could be a teacher; I taught at the UPC (Universitat Politecnica de Catalunya) for several years and enjoyed it a lot. Also, every year, I am one of the advisors of a student team that attends the Student Cluster Competition. We train them for six months in the basics of HPC and accompany them to the ISC conference, where the onsite competition takes place.
For aspiring scientists or individuals interested in a career in geophysics and HPC research, what advice would you give to inspire and guide them in pursuing this field?
I would give them the advice I provide students from all fields: do what you love; that is the first step. Also, from the HPC field, we can help society in many ways, although, sometimes, it is difficult to see.
Beyond the realm of science, can you share a cultural piece (film, book, podcast) that has significantly influenced your perspective or approach to your work?
Maybe not exactly a cultural piece, but something beyond the realm of science helps me in my work: pottery. In my free time, I like to make pottery. Doing something with my bare hands relaxes me a lot; that way, I can think more clearly.
More of the Meet the Researcher series:
By Aerton Guimarães
From the ChEESE-2P Dissemination Team