I don't understand what your code is trying to do or why you talk about "replacements," but if you want to get a random substring from a longer string, why not just pick a random offset between 0 and the difference between the full string's length and the substring's length? Here's an iterator which does that:

#!/usr/bin/env perl
use 5.010; use warnings; use strict;

my $DNA = "AACCGTTAATGGGCATCGATGCTATGCGAGCT";

sub make_rand_getter {
    my $s = shift;
    my $sl = length $s;
    return sub {
        return substr $s, int rand($sl - $_[0]), $_[0];
    }
}

my $rstring = make_rand_getter($DNA);
say "3-letter string: " . $rstring ->(3);
say "7-letter string: " . $rstring ->(7);
[download]