Re^2: performance of counting keys in a big hash

[space_monk]

A good piece of research, however the possibility exists that the two results are related and generating the first one also pre-generates the second one. I suggest that tests like these are best done in separate methods on separate data just to be sure.

#!/usr/bin/perl

use strict;
use warnings;
use Benchmark qw(:all) ;

my $n = 100000;
our %hash1;
@hash1{1..$n} = 1 x $n;
my $n2 = 100000;
our %hash2;
@hash2{1..$n2} = 1 x $n2;
my $n3 = 1000000;
our %hash3;
@hash3{1..$n3} = 1 x $n3;

cmpthese( 1000, {
    'Method 1' => sub  {
            my @k = keys %hash1;
            my $v = scalar @k;
        }, 
    '100K' => sub {
           my $v = scalar keys %hash2;
        },
   '1 Mill' => sub {
           my $v = scalar keys %hash3;
        }
    }
    );

Results:

                         Rate             Method 1         1 Mill     
+      100K

Method 1               15.0/s                   --          -100%     
+     -100%

1 Mill   999999999999999872/s 6645700000000000000%             --     
+        0%

100K     999999999999999872/s 6645700000000000000%             0%     
+        --

Note result appears to be same for 100k and 1Million entries in hash.

[hdb]

This is a valuable comment, I had not thought of that. I have reversed the order and get the same results. Does this provide enough proof or could there be a caveat as well?

To be really sure one would have to run separate scripts?

[space_monk]

I'm just in the process of tidying up my comment by putting in some code that actually works, as opposed to that I first thought of! :-).

I think reversing the tests is probably enough proof; I don't believe separate scripts would prove anything extra.

I am using Benchmark to compare my results instead of using manual timing.