BB.20.008 – Back to the drawing board: reinventing data processing in fundamental physics
Research in particle and gravitational wave physics requires petabytes of data to be efficiently processed by large numbers of bespoke algorithms on limited resources. These algorithms are often designed and implemented for a specific research goal, by researchers themselves, and optimized ad-hoc based on a given researcher’s (limited) knowledge of performance optimization. In addition, algorithms are often only suitable for execution on traditional CPUs, whose improvement in performance lags behind, in recent years, compared to that of accelerator architectures such as GPGPUs. In the absence of accelerators, performance bottlenecks are mitigated by adding more CPU-like resources, or simply waiting longer for results. However, the large increase in data volume, brought by current and next-generation experiments makes both strategies untenable. Thus, to expand particle physics and gravitational wave research, we must develop advanced methods and tools to design accelerator-ready, hardware-agnostic algorithms, and further implement them efficiently on actual state-of-the-art machines. This is the only approach that enables these algorithms to keep pace and make good use of the huge and rapid advancements in computer architecture. Specifically, in this project, we will focus on the design and development of methods and tools for future-proofing data processing in fundamental physics research which requires the combined expertise of both computer scientists and physics researchers. Our goals are to accelerate two science cases: flavour physics with the LHCb experiment at CERN and searches for new types of binary inspirals with the Virgo/Ligo/KAGRA gravitational wave data, which will widen the window on the smallest and largest scales that can be observed in the universe.
algorithms, compute accelerators, Gravitational waves physics, Particle Physics
University of Amsterdam (UvA), Utrecht University (UU), Vrije Universiteit Amsterdam (VU)
|Name||Dr. R. (Roel) Aaij|