There are directories like:
/eos/opendata/cms/dataset-semantics/NanoAODSIM/
with an entry per run. Currently, there are around 21k entries. It would be nice to have these directories spread a bit more, instead of all on a single entry. One possibility would be to make a subdirectory for each 1k, rounding down the run number to the closest thousand. So, for instance:
/eos/opendata/cms/dataset-semantics/NanoAODSIM/34607
would become
/eos/opendata/cms/dataset-semantics/NanoAODSIM/34000/34607