While I'm looking into the possibilities provided by the other posters, these tips don't require more advanced knowledge than I have ;-).

Thank you for warning me about the anonymous array overhead. Will change that immediately.

Furthermore, I'm on little endian, so the string comparison should work just fine. However, I don't want to sort on the remainder of the string as it is now. Will run a test first, and post the results here.

Thanks a lot,

Jeroen
"We are not alone"(FZ)

Update: Must have made some difference in memory usage, but it's still way too much. Made a rough calculation of the memory usage with system monitor (gnome) and this little script:

#!/usr/bin/perl

my $size=1E6;
my $f = 2;
print "Increasing memory print each step by a factor $f\n\n";
while(1){
  print "\tCreating array of $size items...Press enter to continue....
+";
  my $b = <STDIN>;
  my @a=1..$size;
  print "...done. Press enter to continue.";
  $b = <STDIN>;
  chomp $b;
  $b=~ /[qxQX]/ and exit;
  $size *= $f;
}
[download]

It's about 44 megabytes for every million items in an array. No way my 12 M records are going to fit in physical memory. I'm back to Radixsort and alike.