If the files in question are of any significant size, sort the two files by the same key. Then, you can dispense with the use of a hash (and all of the virtual-memory it consumes) altogether.

Important: use a disk-based sort.

As "the COBOL-jockeys of yore" well-knew (and as the present ones still know), sorting a large file is an unexpectedly-fast operation. If you know that your inputs are sorted, then:

• Any occurrences of "the same key" will always occur together.
• When the key-value changes, you must be at the end of all occurrences of the preceding key-value and at the beginning of all occurrences of the new one.
• If any gaps exist, you know without searching that there are no key-values anywhere within that gap.

Although we might not routinely think of "main memory" as "a disk file," that is, in fact, what it is. A page-fault is expensive even if it does not result in physical I/O. If you are dealing with more than, say, a few thousand records, the amount of speed gained can be quite startling. (As in, hundreds of times faster.)

Also... don't let "tie" beguile you. A "tied hash" is not a hash. A convenient metaphor it may be, but a true hash it is not.