Interesting. I will have to look at the Radix sort more in depth.
I was thinking of using a modified heap-sort, because that way if I want, I can only grab the top million or so lines. But I will have to look at this Radix sort, it might work really well.
again, thanks. I will let you know how it turns out, once I finish the script.

I thought I'd have a play, so here's my script - it will sort a file 'NUMBERS.DAT' containing lots of 8 digit decimal numbers into 'SORTED.DAT'

For interest's sake I've left all the temporary files in, and it assumes the input file is zero padded so all numbers are 8 digits, no more, no less.

It sorted a 20,000,000 line file (200MB) in 340 seconds (just over 5 minutes) on my PC never using more than 2MB RAM (well, it probably used more memory indirectly due to disk caching)

You would be able to get it quicker by making the 'radix' 100 instead of 10, it'd probably be twice as quick (heck, if you make the radix 100000000, it'll probably sort it in about 90 seconds, but you could run out of file handles ;-) )

$starttime = time;
printf ("start - %d\n",$starttime);
$pref = "";
for ($j = 0; $j < 10; $j++)
{
    unlink "-$j";
}

sortfile("numbers.dat", $pref, 7);

for ($i = 6; $i >= 0; $i--)
{
    for ($j = 0; $j < 10; $j++)
    {
        unlink "$pref-A-$j";
    }

    for ($j = 0; $j < 10; $j++)
    {
        sortfile("$pref-$j", "$pref-A", $i);
    }
    $pref .= "-A";
}

printf ("end sort - %d (%d)\n",time, time - $starttime);
open (FILE, ">sorted.dat") || die;
for ($i = 0; $i < 10; $i++)
{
    open(IN, "$pref-$i") || die;
    while(<IN>)
    {
        print FILE $_;
    }
    close(IN);
}
close(FILE);

printf ("end - %d (%d)\n",time, time - $starttime);

sub sortfile
{
    my ($source, $pref, $offset) = @_;
    my $i, @fh;
    printf("$offset $pref $source - %d\n", time - $starttime);
    
    open(FILE, $source) || die;
    for ($i = 0; $i < 10; $i++)
    {
        open($fh[$i], ">>$pref-$i") || die;
    }
    
    while(<FILE>)
    {
        $f = $fh[substr($_, $offset, 1)];
        print $f $_;
    }
    
    close(FILE);
    
    for ($i = 0; $i < 10; $i++)
    {
        close($fh[$i]);
    }
}
[download]

(PS - it's actually quite a simple algorithm to understand as well, even I can follow it ;-) )

The algorithm will be able to sort much bigger files in linear time, and without using any more memory

You might like to try this. It sorts an 80MB file in around 5 mins and consumes less than 10 MB of ram.

I will also handle sorting files up to the 4 GB filesystem limit using less than 50 MB of ram, though it will obviously run somewhat more slowly. Given more information about the form of the records, it would be possible to tailor the algorithm to speed the processing.

#! perl -slw
use strict;

open IN, '+<', $ARGV[0] or die $!;
my @splits;

my $pos = 0;
while( <IN> ) {
    $splits[ unpack 'n', substr( $_, 0, 2) ] .= 
          pack( 'V', $pos ) . substr $_, 2, 4;
    $pos = tell IN;
}

@splits = grep $_, @splits;

my $n;
for my $split ( @splits ) {
    $split = join'', map{ substr $_, 0, 4 } sort{
        my( $as, $at, $bs, $bt ) = ( unpack( 'VA4', $a ), unpack( 'VA4
+', $b ) );
        $at cmp $bt
        ||
        do{ seek IN, $as, 0; scalar <IN> }
        cmp
        do{ seek IN, $bs, 0; scalar <IN> }
    } unpack '(A8)*', $split;
}

for my $split ( @splits ) {
    printf do{ seek IN, $_, 0; scalar <IN> } for unpack 'V*', $split;
}
[download]