SCIENCE (aka benchmarking)

Recently we were doing something at work where we needed to get to a location deep in an HoH. We already had a solution that worked alright, but it was copy pasted in a couple places, it wasn’t tested, and it wasn’t documented. So I looked around on CPAN and found Hash::Path. It did exactly what we wanted, but the code was recursive instead of iterative (like our solution.) Because we weren’t going too deep I just installed it and figured I’d look at the actual differences later.

Well, I think last week I felt the urge to see what the difference actually was, empirically speaking. The following is my test case:

#!perl
use strict;
use warnings;
use Time::HiRes 'gettimeofday';
use Hash::Path;
use feature ':5.10';

sub generate_giant_thing {
   my $items = shift;
   my $top_level_data_structure = {};
   my $current = $top_level_data_structure;
   for (0..( $items - 1 )) {
      $current->{"f$_"} = {};
      $current = $current->{"f$_"};
   }
   $current->{"f$items"} = 1;
   return ($top_level_data_structure, [ map "f$_", (0..$items) ]);
}
my ($foo,$path) = generate_giant_thing(500);

sub our_path {
   my $data_set = shift;
   my @hash_keys = @_;
   my $levels = scalar @hash_keys;
   my $return_value =  $data_set->{$hash_keys[0]};
   for (1..($levels - 1)) {
      $return_value = $return_value->{$hash_keys[$_]};
   }
   return $return_value;
}
{
   my $before = gettimeofday;
   say our_path($foo, @{$path});
   my $after = gettimeofday;
   warn 'Our Time: '.sprintf('%0.3f', $after - $before).' seconds';
}

{
   my $before = gettimeofday;
   say Hash::Path->get($foo, @{$path});
   my $after = gettimeofday;
   warn 'HP Time: '.sprintf('%0.3f', $after - $before).' seconds';
}

Ours stayed pretty close to 0.001 seconds, whereas the other version went a little slower (I think up to like, .010 s) but ran out of stack before I could test much deeper. So I put the testcase on RT in the hopes that the developer checked his email. He does and he updated the module just a couple days later! Pretty cool, huh?

Posted Fri, Oct 2, 2009

---

If you're interested in being notified when new posts are published, you can subscribe here; you'll get an email once a week at the most.