I was playing around with the falcon p385 patch to see if it’s any faster than some of the more recent MRI rubies. TL;DR version: looks like p192 is faster than p385 of any type or tweak.

Here’s how to get a p385 Ruby version patched with funny falcon’s performance patches using RVM.

mkdir ~/.rvm/patches/ruby/1.9.3/p385
curl https://github.com/funny-falcon/ruby/compare/p385...p385_falcon.diff > \
$rvm_path/patches/ruby/1.9.3/p385/falcon.patch
rvm install 1.9.3-p385 -n perf --patch falcon

Then rvm use 1.9.3-p385-perf or set it as your global ruby.

Test Setup

The following benchmarks were run on an i7 server with a RAID5 array. The disk is slow (lack of large cache) but the benchmarks were run on the same box so it should compare apples-to-apples.

From here on out, here are the definitions for the Ruby versions.

p194 = 1.9.3p194 default
p385 = 1.9.3p385 default
falcon = 1.9.3p385 with the above falcon diff patch applied
gcc_tweak = the falcon patches with GCC compile flags tweaked.

So what are these GCC tweaks? Explicitly setting the CFLAGS for your machine’s CPU type and recompiling ruby with the Falcon patches applied.

Micro and Macro Benchmarks

I used the ruby-benchmark-suite to run these tests.

Here are some example results. I can’t list them all. There are over 100 benchmarks. These are results for the mean times in seconds.

</p> </p> </p> </p> </p> </p> </p> </p> </p> </p> </p> </p> </p> </p> </p> </tr> </tr>

test	1.9.3p194	1.9.3p385	falcon	gcc_tweak
bm_sudoku.rb	1.379112226	1.598182153	1.495923579	1.526717563
bm_open_many_files.rb	0.175602996	0.197096826	0.197673286	0.194135045
... etc etc

Here’s the winner summary for mean times. This is the number of times the ruby version was the fastest for a particular benchmark. 1.9.3p194 - 64 wins 1.9.3p385 - 29 wins 1.9.3p385-falcon with GCC tweaks - 10 wins 1.9.3p385-falcon 5 wins

Boot time and IO

Timing rails boot time is a bit more important to me. If you want to know how to really save “rails boot time” see the DAS screencast on not loading Rails at all.

Even when using domain objects and lib tricks, it’s nice to have Rails and all I/O boot fast. The main thing that funny falcon’s patches do is speed up requires and I/O.

So let’s benchmark booting a Rails app. $ time bundle exec rake environment

</p> </p> </p>

Ruby Version	Seconds
p385 patched with falcon and GCC tweaks	2.481 total
p374 defaults	3.336 total
2.0.0-rc2	2.613 total

In conclusion, p194 looks faster on the macro and micro benchmarks but Falcon patches boot Rails faster.