If using Perl's hashes means you are running out of memory (or into swapping) then using a file based hash (eg. BerkeleyDB ) will get you around that problem, but will be slow. Slower to access (compared to memory-based hashes); much slower to construct.

Depending what you are doing with the hash, there are sometimes alternatives that provide some functionality similar to a hash that are more compact that can avoid moving to disk-based. But you'd need to describe the data, and the use you are making of the hash.