The raw output files are organized along the Peano-Hilbert curve which could be non trivial to manipulate. Moreover these files contain a lot of double-precision and not very useful information.
We have therefore developped two MPI codes to make the data lighter and user-friender: Parallel Slicer and Parallel FoF.These two codes have been succesfully tested up to 32768 Blue Gene/P cores.
Parallel Slicer reads RAMSES data (Fields, Halos and Lightcones) and redistributes them in cubic subvolumes. Each file corresponds to a given region, which allows to load only the region of interest. The files contain the coordinate of the region, particles positions, velocities (simple precision) and identities (long).
This code is a parallel halo finder based on a Friend-of Friend algorithm. The principle is the following. Each processor loads particles corresponding to a given cubic region (see Parallel Slicer above) and runs FoF locally. When some particles are close to the edge of a cube, the code checks if the halo extends in neighbouring cubes. This is done iteratively until all halos have been merged. Particles are then sorted halo per halo and each processor writes a file with all the particles belonging to a given number of halo. Each file contains the identity of the halo (long), particles positions, velocities (simple precision) and identities (long). This makes any subsequent analysis straightforward. This code has been intensively used in Courtin et al, 2011 and has successfully participated to the Halo Finder Comparison Project (Knebe et al, 2011).