pFoF: a highly scalable halo-finder for large cosmological data sets

We present a parallel implementation of the friends-of-friends algorithm and an innovative technique for reducing complex-shaped data to a user-friendly format. This code, named pFoF, contains an optimized post-processing workflow that reduces the input data coming from gravitational codes, arranges them in a user-friendly format and detects groups of particles using percolation and merging methods. The pFoF code also allows for detecting structures in sub- or non-cubic volumes of the comoving box. In addition, the code offers the possibility of performing new halo-findings with a lower percolation factor, useful for more complex analysis. In this paper, we give standard test results and show performance diagnostics to stress the robustness of pFoF. This code has been extensively tested up to 32768 MPI processes and has proved to be highly scalable with an efficiency of more than 75%. It has been used for analysing the Dark Energy Universe Simulation: Full Universe Runs (DEUS-FUR) project, the first cosmological simulations of the entire observable Universe, modelled with more than half a trillion dark matter particles