Maybe (maybe), you can get away by splitting all search terms into non-digits and digits, and sorting on that. This means that all items (in a STB population) have the same number of items, and that nobody mixes letters and numbers within the item:

#!/usr/bin/perl

use strict;
use warnings;

use Test::More tests => 6;

sub natural_string_cmp {
    my ($left, $right) = @_;

    my @left  = split /\b|(?<=[A-Za-z])(?=\d)|(?<=\d)(?=[A-Za-z])/, $l
+eft;
    my @right = split /\b|(?<=[A-Za-z])(?=\d)|(?<=\d)(?=[A-Za-z])/, $r
+ight;

    #use Data::Dumper;
    #warn Dumper \@left;
    #warn Dumper \@right;

    # Now, reconstruct a string for each item that can simply be compa
+red directly.
    # For this, we zero-left-pad all numbers and extend all strings wi
+th \0
    # For simplicity, I assume no number longer than 18 digits and no 
+string longer than 10 characters
    my $l = join "", map { /\d/ ? sprintf '%018d', $_ : substr($_."\0\
+0\0\0\0\0\0\0\0\0",0,10) } @left;
    my $r = join "", map { /\d/ ? sprintf '%018d', $_ : substr($_."\0\
+0\0\0\0\0\0\0\0\0",0,10) } @right;

    #warn $l;
    #warn $r;

    return $l cmp $r
}

is natural_string_cmp('Rack1-Unit1', 'Rack1-Unit1'), 0, "Identity";
is natural_string_cmp('Rack1-Unit1', 'Rack1-Unit2'), -1, "Smaller";
is natural_string_cmp('Rack1-Unit2', 'Rack3-Unit1'), -1, "Smaller";
is natural_string_cmp('Rack3-Unit1', 'Rack1-Unit1'),  1, "Larger";


is natural_string_cmp('R1U1', 'R1U1'), 0, "Identity";
is natural_string_cmp('R1U2', 'R3U1'), -1, "Smaller";
is natural_string_cmp('R3U1', 'R10U1'), -1, "Smaller";
is natural_string_cmp('R10U1', 'R3U1'),  1, "Larger";


#R1-U1
#R1-U10
#R2-U2
#R3-U13
#R10-U1
#R10-U5
#R10-U11
#
[download]