use strict;

open( CDR, $CDR ) || die "Can't read $CDR\n";
open( DUP, ">$CDR.dup" ) || die "Can't write $CDR.dup\n";
open( UNQ, ">$CDR.unq" ) || die "Can't write $CDR.unq\n";
my %seen;

while (<CDR>) {
    if ( exists( $seen{$_} )) {
        print DUP;
    }
    else {
        $seen{$_}++;
        print UNQ;
    }
}
[download]

Having all the unique data lines in a hash in memory shouldn't be a problem for the size of files you mentioned.

Note that the numbe of lines in the "dup" file plus the number of lines in the "unq" file should sum to the number of lines in the input file.

This is untested, but it ought to do the trick, and it has the added bonus of not having to read 4 megs of text into memory all at once.

#! /usr/bin/perl

use strict ;
use warnings ;
$|++ ;

open(CDR, $CDR);
open(SORTED_CDR, ">$SORTED_CDR");
open DUPS, 'dup_records.txt' ;
#@ARRAY=<CDR>;
#my @unique = do {my %h; grep {!$h {$_} ++} @ARRAY};
my %seen = () ;
while ( <CDR> )
{
    if ( $seen{$_}++ )
    {
        print DUPS $_ ;
    }
    else
    {
        print SORTED_CDR $_ ;
    }
}
#print SORTED_CDR @unique;

exit ;

__END__
[download]

'scuse if I got this wrong--I have used a *nix system for several years--but wouldn't this work for you(on a *nix system)?

uniq -cd sorted_cdr >duplicates
uniq -u  sorted_cdr >unique_cdr
[download]

Of course, a perl solution will be more portable if that's a req.

sorry BrowserUK, I should have mentioned it was not a *nix system.