I would think that doing as much of the work as possible in the SQL query, as suggested by other Monks, makes most sense. I have not done enough SQL to usefully contribute there. Looking at the documentation for DBI I saw the fetchall_arrayref method. If going the simple query route, you could use it to populate a hash keyed by service_id with the results of the query sorted by descending distance. Thus the shortest distance row for each service would be assigned to the hash last, overwriting any longer distances. I don't have DBI available (or a database for that matter) so have made up a pretend routine that returns a ref. to an AoA and I added a little more pretend data.

use strict;
use warnings;

use Data::Dumper;

sub pretendFetch
{
    my $cannedFetch = [
       [6200249230, 2, 247.695377429616],
       [6200200581, 1, 131],
       [6200249220, 2, 394.81641303269],
       [6200249240, 2, 148.32734070292],
       [6200249269, 1, 127.84583439484],
       [6200249250, 2, 394.507287638644],
       [6200248437, 1, 234.38349845484],
       ];
    return $cannedFetch;
}

my %service_id =
   map  { $_->[1] => {
                        stop_reference => $_->[0],
                        distance => $_->[2]
                     }
        }
   sort { $b->[2] <=> $a->[2] }
   @{ pretendFetch() };

print Data::Dumper->Dump([\%service_id], [q{*service_id}]);
[download]

Which produces

%service_id = (
                '1' => {
                         'distance' => '127.84583439484',
                         'stop_reference' => '6200249269'
                       },
                '2' => {
                         'distance' => '148.32734070292',
                         'stop_reference' => '6200249240'
                       }
              );
[download]

I hope this is of interest.

Cheers,

JohnGG