From: Pit Capitain <pit@...>
Date: 2007-01-31T20:08:30+09:00
Subject: Re: Benchmark results

Shot (Piotr Szotkowski) schrieb:
> Ok, I guess I have a question on what does exactly Benchmark results
> mean. Doesn�t the below mean hash_map_sort is actually a bit slower than
> hash_sort?

This is what I would expect. The advantage of Mauricio's solution is 
that you can use it even if the Set elements are not comparable to each 
other. My proposal was meant for your use case with integers as Set 
elements.

> srand 1979
> array = (0..1_000_000).sort_by {rand}
> 
> Benchmark.bmbm do |b|
>   b.report 'hash_sort' do
>     Block.new(array).hash_sort
>   end
>   b.report 'hash_map_sort' do
>     Block.new(array).hash_map_sort
>   end
> end

With this benchmark, you also measure the time needed to construct the 
Block instances. If you want to compare the two #hash methods, you 
should change your benchmark to something like:

   block = Block.new(array)

   Benchmark.bmbm do |b|
     b.report 'hash_sort' do
       block.hash_sort
     end
     b.report 'hash_map_sort' do
       block.hash_map_sort
     end
   end

 From your benchmark, you can see that it takes a lot of time to compute 
the hash values. Will your sets contain 1_000_000 elements? How often 
will they change? How are they created? Maybe you need to cache the hash 
values or implement a different hash algorithm.

Regards,
Pit