Green computing for particle physics

Climate change-related events are starting to deeply concern modern societies and scientific communities all over the world, who agree that human activities have contributed to this phenomenon and actions will be required to curb our current and future impact. Among these activities, due to its scale and pervasiveness, computing has been recognized having a relevant footprint on the environment. Several scientific computing communities are therefore taking action to acknowledge their computational footprint and reduce it, aiming at guaranteeing an “optimal” energy consumption-per-unit-of-knowledge obtained. HEP physicists, given the upcoming HL-LHC phase of the LHC experiment which will scale computing to exascale, have recently started taking action aimed at evaluating the footprint of current activities to ensure the sustainability of future scientific efforts at LHC. This thesis, in continuity with the aforementioned collective effort and with the goal of helping physicists to easily acknowledge their computational and energy footprint, presents the containerized prototype of an energy footprint monitoring software of arbitrary computing tasks running on common computing platforms. The software leverages basic properties of Linux systems to seamlessly perform the estimation of the energy consumption of tasks without including other processes running on the machine in the estimation. In order to test this software prototype, benchmark CMS workflows related to event generation and simulation, digitization, and reconstruction have been analyzed. The obtained results are reported and used for further analysis aimed at finding margins of sustainability improvements. We show that optimal working points can be extracted and used to prompt a less energy-eager job submission on older platforms, in exchange of a moderate increase in the running time of the job.