This gem is an exercise of style, devised to investigate:
- the best practices in the development of Ruby gems containing C code (traditional MRI C extension interface does not work in JRuby and should thus be ditched in favor of more portable solutions);
- the potential performance gains, for scientific computing purposes, offered by state-of-the-art Pseudo Random Number Generators (PRNGs) over the old school PRNG based on the Mersenne Twister algorithm implemented in both MRI and JRuby.
As a result, I strongly recommend not to use this code for any real-life applications. Indeed, at the moment of this writing (June 14th, 2020), the code in this gem is slower than MRI's and JRuby's default PRNG implementation. (See section Benchmark below for more information.)
Add this line to your application's Gemfile:
gem 'fastrandom'And then execute:
$ bundle
Or install it yourself as:
$ gem install fastrandom
Here are the results obtained when executing the bin/benchmark script with JRuby 9.2.11.1 on a 2016 Dell XPS running Linux Arch 64-bit (kernel 5.6.13):
jruby-9.2.11.1
Rehearsal ----------------------------------------------
fastrandom 1.770000 0.060000 1.830000 ( 0.967270)
random 0.700000 0.030000 0.730000 ( 0.377934)
------------------------------------- total: 2.560000sec
user system total real
fastrandom 0.730000 0.050000 0.780000 ( 0.726169)
random 0.320000 0.020000 0.340000 ( 0.332982)
The benchmark compares fastrandom with the default random number generator, indicated as random, of the Ruby system library.
And here are the results with MRI 2.7.1 on the same machine:
2.7.1
Rehearsal ----------------------------------------------
fastrandom 0.911116 0.003769 0.914885 ( 0.926847)
random 0.351022 0.009696 0.360718 ( 0.365660)
------------------------------------- total: 1.275603sec
user system total real
fastrandom 0.919464 0.003296 0.922760 ( 0.934760)
random 0.354044 0.003121 0.357165 ( 0.361642)
Finally, here are the results with TruffleRuby 20.1.0 (installed through rbenv) on the same machine:
Rehearsal ----------------------------------------------
fastrandom 5.863129 0.022847 5.885976 ( 4.281766)
random 1.507686 0.003397 1.511083 ( 1.018081)
------------------------------------- total: 7.397059sec
user system total real
fastrandom 2.870086 0.000061 2.870147 ( 2.885172)
random 0.770102 0.000004 0.770106 ( 0.766706)
truffleruby-20.1.0
As you can see, despite leveraging the xoroshiro128+ algorithm, which is deemed to be significantly faster than the old and venerable Mersenne Twister used by the default PRNG of both MRI and JRuby, fastrandom is overall noticeably slower than the default PRNG. I haven't been able to pinpoint the reason yet, but I speculate that the overhead brought by the FFI interface dominates the performance gains brought by xoroshiro.
Bug reports and pull requests are welcome on GitHub at https://github.com/mtortonesi/fastrandom.
The gem is available as open source under the terms of the MIT License.