Use 'd', it will handle integers upto 2⁵³ without loss of accuracy which means your set for files upto ~8,000,000 GB, which should give you room for a little future growth. Be aware that your index will require more memory.

It was pointed out that whilst using 'd' to pack your offsets will work, when it comes to doing the sorting, you would have to unpack the values and do a numeric comparison as sorting numeric data in it's pack'd binary for doesn't work for packed floats/doubles.

Having previously extolled the virtues of sorting numeric data in it's binary form, that doesn't sit right, so here are a couple of routines that will pack and unpack a FP value < 2**53 to an 8-byte binary form that is sortable. along with some rudimentary tests:

#! perl -slw
use strict;
use Math::Random::MT qw[ rand ];

sub ftob64 {
    return pack 'NN', int( $_[ 0 ] / 2**32 ), int( $_[ 0 ] % 2**32 );
}

sub b64tof {
    my( $hi, $lo ) = unpack 'NN', $_[ 0 ];
    return $hi * 2**32 + $lo;
}

for ( 1 .. 1e6 ) {
    my $test = int( rand 2**53 );
    my $b64 = ftob64 $test;
    my $float = b64tof $b64;
    if( abs( $test - $float ) > 1e-15 ) {
        printf "%31.f v %31.f => diff %31.31f\n", 
            $test, $float, abs( $test - $float );
    }
}

my @randomBin = map{ ftob64 int rand 2**53 } 1 .. 1e6;

my @sortedBin = sort @randomBin;

my @sortedN = map{ b64tof $_ } @sortedBin;

$sortedN[ $_ ] > $sortedN[ $_ + 1 ] 
    and die "Error: $_ : $sortedN[ $_ ] > $sortedN[ $_ + 1 ]"
    for 0 .. $#sortedN - 1;
[download]