So, does the act of putting a hash into DBM::Deep cause the keys to be sorted (that would make sense for an efficient on-disk format and would be a very welcome feature to me) or are you suggesting that using perl's sort on an array from DBM::Deep would efficiently swap parts in and out of memory such that the sorting would work reasonably fast and also not require that much memory? (update: I guess the latter would also require storing the sorted results into a DBM::Deep array, perhaps the same one)