Writing Fast Ruby

Transcript

1. View Slide

2. View Slide

3. View Slide

4. View Slide

5. Levels of Optimization
   Design
   Source
   Build
   Compile
   Runtime
   

   View Slide

6. Levels of Optimization
   Design
   Source
   Build
   Compile
   Runtime
   Architecture and algorithms (e.g. n + 1 queries)
   Writing fast Ruby
   Setting build flags (e.g. ./configure)
   mrbc, jrubyc, rbx compile
   Thanks Matz & Koichi (e.g. RUBY_GC_MALLOC_LIMIT)
   

   View Slide

7. View Slide

8. Benchmark
   require ‘benchmark'
   n = 50
   Benchmark.bm do |x|
   x.report { n.times { fast } }
   x.report { n.times { slow } }
   end
   

   View Slide

9. Benchmark
   require ‘benchmark'
   n = 50_000
   Benchmark.bm do |x|
   x.report { n.times { fast } }
   x.report { n.times { slow } }
   end
   

   View Slide

10. Benchmark IPS
    require 'benchmark/ips'
    Benchmark.ips do |x|
    x.report('fast') { fast }
    x.report('slow') { slow }
    end
    

    View Slide

11. Goals
    Source
    Significant
    Easy
    Happy
    Optimize at the code level
    At least 12% improvement
    Code should be easier to read
    High quality Ruby
    

    View Slide

12. Proc#call versus yield
    def slow(&block)
    block.call
    end
    def fast
    yield
    end
    

    View Slide

13. Results
    slow 950950.6 (±14.0%) i/s
    fast 5508226.3 (±15.5%) i/s
    Over 5X faster!
    

    View Slide

14. Proc#call versus yield
    def slow
    Proc.new.call
    end
    def fast
    yield
    end
    

    View Slide

15. Block versus Symbol#to_proc
    (1..100).map { |i| i.to_s }
    (1..100).map(&:to_s)
    

    View Slide

16. Results
    slow 47524.3 (±7.6%) i/s
    fast 56823.2 (±7.2%) i/s
    20% faster!
    

    View Slide

17. View Slide

18. View Slide

19. Enumerable#map and Array#flatten
    versus Enumerable#flat_map
    enum.map do
    # do something
    end.flatten(1)
    enum.flat_map do
    # do something
    end
    

    View Slide

20. Results
    slow 12348.2 (±9.0%) i/s
    fast 56647.8 (±7.2%) i/s
    Over 4.5X faster!
    

    View Slide

21. View Slide

22. View Slide

23. Enumerable#reverse and Enumerable#each

    versus Enumerable#reverse_each
    enum.reverse.each do
    # do something
    end
    enum.reverse_each do
    # do something
    end
    

    View Slide

24. Results
    slow 156173.2 (±9.2%) i/s
    fast 182859.3 (±7.8%) i/s
    17% faster!
    

    View Slide

25. View Slide

26. Hash#keys and Enumerable#each

    versus Hash#each_key
    hash.keys.each do |k|
    # do something
    end
    hash.each_key do |k|
    # do something
    end
    

    View Slide

27. Results
    slow 34702.1 (±9.8%) i/s
    fast 46103.3 (±8.1%) i/s
    Over 33% faster!
    

    View Slide

28. View Slide

29. View Slide

30. Array#shuffle and Array#first

    versus Array#sample
    array.shuffle.first array.sample
    

    View Slide

31. Results
    slow 324806.7 (±8.1%) i/s
    fast 5069719.9 (±9.5%) i/s
    Over 15X faster!
    

    View Slide

32. View Slide

33. View Slide

34. Hash#merge versus Hash#merge!
    enum.inject({}) do |h, e|
    h.merge(e => e)
    end
    enum.inject({}) do |h, e|
    h.merge!(e => e)
    end
    

    View Slide

35. Results
    slow 33572.0 (±3.3%) i/s
    fast 106473.0 (±3.4%) i/s
    Over 3X faster!
    

    View Slide

36. Hash#merge! versus Hash#[]=
    enum.each_with_object({}) do |e, h|
    h.merge!(e => e)
    end
    enum.each_with_object({}) do |e, h|
    h[e] = e
    end
    

    View Slide

37. Results
    slow 99331.3 (±2.2%) i/s
    fast 217944.4 (±5.2%) i/s
    Over 2X faster!
    

    View Slide

38. Hash#fetch versus Hash#fetch with block
    {:ruby => :conf}.fetch(:ruby, (0..9).to_a)
    {:ruby => :conf}.fetch(:ruby) { (0..9).to_a }
    

    View Slide

39. Results
    slow 712384.2 (±3.8%) i/s
    fast 1590417.4 (±4.1%) i/s
    Over 2X faster!
    

    View Slide

40. String#gsub versus String#sub
    ‘http://rubyconf.pt/'.gsub(‘http://', 'https://')
    'http://rubyconf.pt/'.sub('http://', 'https://')
    

    View Slide

41. Results
    slow 404148.2 (±4.6%) i/s
    fast 602661.1 (±3.4%) i/s
    50% faster!
    

    View Slide

42. String#gsub versus String#tr
    'slug from title'.gsub(' ', '_')
    'slug from title'.tr(' ', '_')
    

    View Slide

43. Results
    slow 311878.2 (±3.5%) i/s
    fast 1573891.1 (±4.6%) i/s
    Over 5X faster!
    

    View Slide

44. Parallel versus sequential assignment
    a, b = 1, 2 a = 1
    b = 2
    

    View Slide

45. Results
    slow 5821588.3 (±6.0%) i/s
    fast 8010420.3 (±5.5%) i/s
    40% faster!
    

    View Slide

46. Using exceptions for control flow
    begin
    ruby.conf
    rescue NoMethodError
    'conf'
    end
    if ruby.respond_to?(:conf)
    ruby.conf
    else
    'conf'
    end
    

    View Slide

47. Results
    slow 328886.4 (±5.0%) i/s
    fast 3348327.9 (±9.0%) i/s
    Over 10X faster!
    

    View Slide

48. Using throw/catch for control flow
    begin
    ruby.conf
    rescue NoMethodError
    'conf'
    end
    catch(:ruby) do
    if ruby.respond_to?(:conf)
    ruby.conf
    else
    throw(:ruby, ‘conf’)
    end
    end
    

    View Slide

49. Results
    slow 252474.2 (±8.4%) i/s
    fast 1411916.6 (±7.2%) i/s
    Over 5X faster!
    

    View Slide

50. The Future
    

    View Slide

51. View Slide

52. View Slide

53. View Slide

54. View Slide

55. View Slide

56. Special Thanks
    Aaron Patterson
    Ruby Rogues Parley
    Sam Saffron
    Aman Gupta
    Don Knuth
    Yukihiro Matsumoto
    Koichi Sasada
    RubyConf Portugal
    

    View Slide

57. View Slide

58. View Slide