It should be noted that at no point does the code sort 14M arrays. The variable @rows actually holds the fields for the current line. The code sorts, for every column, the numbers of occurrences of different values. While a column could hold 14M different values in 14M lines, this is not the case here: With 14M lines of 1400 fields each, and 62GB in total, the average column has a data width of two bytes. You can only cram so much different values into two bytes (especially if it's text) - that's several orders of magnitude away from 14M and should fit into memory quite easily.

The sorting problem in the code is, as has been pointed out, that the sorting is done 14M times instead of once.