Performance optimization ofthe online data processing softwareof a high-energy physics experiment : Performance optimization ofthe online data processing softwareof a high-energy physics experiment

Abstract: The LHCb experiment probes the differences between matter and anti-matter by examining particle collisions. Like any modern high energy physics experiment, LHCbrelies on a complex hardware and software infrastructure to collect and analyze the data generated from particle collisions. To filter out unimportant data before writing it to permanent storage, particle collision events have to be processed in real-time which requires a lot of computing power. This thesis focuses on performance optimizations of several parts of the real-timedata processing software: i) one of the particle path reconstruction steps; ii) theparticle path refining step; iii) the data structures used by the real-time reconstructionalgorithms. The thesis investigates and employs techniques such as vectorization, cache-friendly memory structures, microarchitecture analysis, and memory allocation optimizations. The resulting performance-optimized code uses today's many-core, data-parallel,superscalar processors to their full potential in order to meet the performance demands of the experiment. The thesis results show that the reconstruction step got3 times faster, the refinement step got 2 times faster and the changes to the datamodel allowed vectorization of most reconstruction algorithms.