I would try and use the system's sort(1) first as it's already optimized for this kind of stuff. 500M 32-bit integers would barely fit your memory if you put them in a straight C-style array with none of Perl's overhead, so it's probably a good idea to do a disk-based mergesort. Split the file into half a dozen or so parts, run sort on them individually and then use sort once more with -m to merge the results.
If you really need to do it in Perl, you could have a look at File::Sort but I have no idea whether it works well.
