Theory:

A hash in perl uses at least 32 bits for the bucket index, so that would translate to 2Gig buckets, so a perfectly filled hash could contain 2Gig elements (if not perfectly filled, it could be a lot more or a lot less). Also, depending on your machine, it could use more bits for the index.

Practice:

This code:

#!/usr/bin/perl -w
use strict;

$|=1;
my @nodes;

for (1 .. 3000) {
    push @nodes,{ number => $_ };
    print "\r$_" unless $_ % 1000;
}

my @edges;
for (0 .. 2500000) {
    push @edges,{ s => \ $nodes[$_ % 3000], d => \ $nodes[3000 - $_ % 
+3000] };
    print "\r$_" unless $_ % 1000;
}

print "\ndone, press enter to exit";
my $dummy = <STDIN>;
[download]

Eats up about 430 Mb on my machine: perl v5.8.0 built for i386-linux-thread-multi.

It's not super-fast, but it still works. However, how much memory Graph::Directed objects will take, and how long operations on the graphs will take is something you'll have to test for yourself.

Joost.

-- 
#!/usr/bin/perl -np
BEGIN{@ARGV=$0}s(^([^=].*)|=)()s;
=Just another perl hacker\
[download]