Over 9000: JRuby in 2015

JRuby 9000 represents the biggest-ever leap forward for JRuby. Not only have we caught up on compatibility (9000 will be 2.2-compatible from release), but we've completely redesigned our JVM-based runtime and have opened our codebase up to the JRuby+Truffle research project from Oracle Labs. The changes we've made will make it easier to keep up with MRI on compatibility and give us the potential to run Ruby as fast as Java or C. The entire Ruby world will change over the next year, and JRuby 9000 will be leading the way. We'll talk about what Ruby's going to look like once JRuby is "over 9000".

Transcript

1. Over 9000 JRuby in 2015
2. Me • Charles Oliver Nutter • headius@headius.com • @headius • Red Hat
4. Ruby on the JVM JVM SUCKS ROFL AbstractMetaRubyImplementationFactoryFactoryImpl I don't like Java so I don't like JRuby LOL applets
5. JRuby is Ruby!
   on the JVM... shhh!
6. Stable: 1.7.19 • Ruby 1.8 and 1.9 support • AST interpreter and JVM bytecode JIT • Easy integration with JVM libraries • No support for MRI C ext
7. Oldest alternative Ruby…around for 4-5 years before any others.
8. Aaron Patterson Aditya Bhardwaj Akinori MUSHA Alan Moore Alex Coles Alex Dowad Alex Tambellini Aliaksei Palkanau Aman Gupta Anders Bengtsson Andreas Woess Andrew Grimm Andrew Kiellor Andy Lindeman Anil Wadghule Anoop Sankar Anthony W. Juckel Antoine Toulme Arne Hormann Arun Agrawal Aslak Hellesøy Atsuhiko Yamanaka Ben Browning Benoit Cerrina Benoit Daloze Bernerd Schaefer Bernhard Urban Bill Dortch Bob Beaty Bob McWhirter Bob Potter Bohuslav Kabrda Brad Heller Brandur Brian Browning Brice Figureau Bruce Adams Bruno Oliveira Chad Fowler Charles Oliver Nutter Charlie Somerville Chris Andrews Chris Heald Chris Jester-Young Chris Price Chris Seaton Chris Sinjakli Chris White Christian Christian Meier Christoffer Sawicki Clayton O'Neill Clayton Wheeler Colin Jones Conrad Irwin Daniel Azuma Daniel Hahn Daniel Lucraft Daniel Luz Daniel Marcotte Daniel Noll Daniel Pittman Dario Bertini Dave Thomas David Calavera David Corbin David Grayson David Hudson David Kellum David Masover David Pollak Deepak Giridharagopal Dennis Ranke Dmitry Ratnikov Don Schwartz Douglas Campos Dwayne Litzenberger Ed Sinjiashvili Edward Anderson Eric Sendelbach Erik Michaels-Ober Frederic Jean Garrett Conaty Gerard Fowley Gino Lucero Greg Mefford Gustav Munkby Gustavo Frederico Temple Pedrosa Heiko W. Rupp Hiro Asari Hironobu Nishikokura Hiroshi Nakamura Hongli Lai (Phusion) Iain Barnett Ian Dees Isaiah Peng Jacob Evans Jake Goulding James Abley James Pickering Jan Arne Petersen Jan Graichen Jan Xie Jason Karns Jason Staten Jason Voegele Javier Jay Jeff Pace Jeff Simpson Jeff Stone Jeremy Evans Jez Ng Joe Joe Kutner Joey Gibson John Croisant John F. Douthat John Firebaugh John Shahid Jon Zeppieri Jonathan Adams Jordan Sissel Jose Rivera Josef Haider Joseph LaFata Josh Ballanco Josh Matthews Joshua Go Juergen Herzog Kamil Bednarz Karol Bucek Kenichi Kamiya Ketan Padegaonkar Kevin Menard Kohsuke Kawaguchi Koichiro Ohba Konstantin Haase Kouhei Sutou Kristian Meier Kubo Takehiro Kyrylo Silin Lars Westergren Lelon Stoldt Leonardo Borges Lin Jen-Shin Loren Segal Luca Simone Lucas Allan Amorim Malte Swart Manish Marcin Mielzynski Marcin Mielżyński Mark McCraw Mark Rada Mark Triggs Mark Warren Martin Harriman Martin Ott Martin Traverso Mateusz Lenik Matjaz Gregoric Matt Hauck Matt Wilbur Matt Wilson Matthew Denner Matthew Kerwin Matthias Grimmer Maximilian Konzack MenTaLguY Micah Martin Michael J. Cohen Michael Klishin Michael Kohl Michal Papis Mike Dalessio NAKAMURA NARUSE, Yui Naoto "Kevin" IMAI TOYODA Nicholas Jefferson Nick Howard Nick Klauer Nick Klauer (a03182) Nick Muerdter Nick Sieger Ola Bini Ole Christian Rynning Olov Lassus Ori Kremer Pablo Varela Patrick Mahoney Patrick Plenefisch Patrick Toomey Paul Brown Paul Mucur Paul Phillips Pekka Enberg Peter Suschlik Peter Vandenabeele Phil Smith Philip Jenvey Pierre-Yves Ritschard Pierrick Rouxel Prathamesh Sonpatki Rajarshi Das Rhett Sutphin Rick Ohnemus Riley Lynch Robert Glaser Robin Dupret Robin Message Rohit Arondekar Ron Dahlgren Ryan Blue Ryan Brown Ryan Fowler Sakumatti Luukkonen Samu Voutilainen Satoru Chinen Scott Blum Scott Clasen Seamus Abshere Sebastian Staudt Sebastien Le Callonnec Seth Wright Shugo Maeda Smit Shah Stefan Huber Stefan Matthias Aust Stephen Bannasch Steven Cook Steven Parkes Subramanya Sastry Syver Enstad Sébastien Le Callonnec Ted Pennings Teemu Theo Theo Hultberg Thomas E Enebo Thomas E. Enebo Thomas Enebo Thomas Wuerthinger Tim Felgentreff Tobi Vollebregt Tobias Crawley Travis Tilley Tristan Hill Unbit Uwe Kubosch Vipul A M Vishnu Gopal Vitor de Lima Vladimir Sizikov Vít Ondruch Wayne Meissner William Thurston Xavier Shay Xb Yoko Harada Yosuke Zach Anker amuino areman arkxu donv elcubo enebo eregon geemus hmalphettes james jc00ke john muhl jonforums josedonizetti kares kiichi kristian lfstad-bren mkristian mohamed peter royal qbproger rdp retnuh rogerdpack rohit ryenus sglee77 simonjsmithuk takeru tduehr the8472 thedarkone timfel tnarik uid41545 unknown wayne williamd wpc yousuke zszugyi
10. But why?
11. Ruby is not perfect.
12. Sometimes, Ruby is the wrong tool.
13. We can make Ruby the right tool for more jobs.
15. Preview: JRuby 9000 • Ruby 2.2 compatible • New optimizing runtime and compiler • Reworked IO, Process, encoding, and more • JRuby 9.0.0.0.pre1 is already out there!
16. Challenges • Concurrency • Language and Libraries • Straight-line Performance • Garbage Collection • Tooling
17. Concurrency
18. True Parallellism Ruby Threads Native Threads Ruby 1.8.7 Ruby 2.0.0 Green Threading CPU Cores in Use JRuby Global Lock Single Thread Real Threading
19. Multicore in MRI 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance Ten-way concurrency * 200MB = 2GB
20. Multicore in MRI 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 100-way concurrency * 200MB = 20GB 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance 200MB MRI Instance
21. Multicore in JRuby 300MB JRuby Instance One instance across 10 threads = 300MB
22. Multicore in JRuby 300MB JRuby Instance One instance across 100 threads = 300MB
23. MRI 1.9.3+, Eight Threads
24. JRuby, Eight Threads
25. Presentation: Bringing Concurrency to Ruby
26. Concurrency Futures
27. Education
28. concurrent-ruby
29. Work with ruby-core on concurrency primitives
30. Mutation testing and monitoring
31. Object ownership and explicit transfer
32. Language and Libraries
33. In MRI, two choices
34. Ruby
35. C
36. In JRuby, you can use…
37. Ruby
38. Java
39. Scala
40. Clojure
41. Javascript
42. Micro Focus Visual Cobol
43. And dozens of others
44. And any language callable with FFI
45. JRuby 1.6 C Exts • Limited support (now disabled) • Removed to a separate repo • If you want it, support it • Some stuff worked...most didn’t
46. Problems • Performance • Data copying to emulate raw structs • Locking to keep C code thread-safe • Multiple JRuby instances in one JVM • No way from C to know which one • Huge API to support
47. Alternatives
48. Java Integration • Call Java (Scala, Clojure, ...) from Ruby • Smart mapping of method names • Type conversions as appropriate • Super easy and fun
49. import javax.swing.JFrame import javax.swing.JLable frame = JFrame.new("Window") label = JLabel.new("Hello") frame.add(label) frame.default_close_operation = JFrame::EXIT_ON_CLOSE frame.pack frame.visible = true
50. Java Native Extensions • Similar to C ext for MRI, but with Java • Fast call protocol...basically free • Same GC for all objects • Have to keep in sync if C version too
51. @JRubyMethod
 public IRubyObject each(ThreadContext context, Block block) {
 if (!block.isGiven()) {
 return enumeratorizeWithSize(context, this, "each", enumLengthFn());
 }
 
 for (int i = 0; i < realLength; i++) {
 block.yield(context, safeArrayRef(values, begin + i));
 } 
 return this;
 }
52. FFI • Ruby API/DSL for calling native code • Came from Rubinius • Runs on all Ruby impls • Solves "access" use case • Works well for coarse-grained calls
53. Ruby FFI example class Timeval < FFI::Struct layout :tv_sec => :ulong, :tv_usec => :ulong end module LibC extend FFI::Library ffi_lib FFI::Library::LIBC attach_function :gettimeofday, [ :pointer, :pointer ], :int end t = Timeval.new LibC.gettimeofday(t.pointer, nil)
54. Language and Library Futures
55. 9000 will match MRI
56. More core in Ruby
57. Eliminating object wrappers
58. Better two-way integration with JVM languages
59. Better FFI tooling in JRuby and JVM
60. Performance
61. Steady Improvement
62. MRI 1.9 ~ 2-3x 1.8
63. MRI 2.1 ~ 2-3x 1.9
64. Faster than most interpreted languages
65. Not even close to compiled/jitted speed
66. JRuby • Faster than MRI on average • Steady improvement • Approaching Java in some areas
67. JVM Over Time 0 6 12 18 24 Java 1.4 Java 5 Java 6 Java 7 JRuby 1.0.3 (bm_red_black_tree.rb)
68. Versus MRI 1.8 0 6 12 18 24 Java 1.4 Java 5 Java 6 Java 7 JRuby 1.0.3 (bm_red_black_tree.rb) MRI 1.8
69. 0 0.55 1.1 1.65 2.2 1.1.6 1.4.0 1.5.6 1.6.8 1.7.0 OpenJDK 8 (bm_red_black_tree.rb)
70. The Good • Method dispatch cheap or free • Method activation cheap or free • Languages inline together (Java, Ruby, Scala…) • Shared resources across languages
71. red/black tree, pure Ruby versus native ruby-2.0.0 + Ruby ruby-2.0.0 + C ext jruby + Ruby Runtime per iteration 0 0.75 1.5 2.25 3 0.29s 0.51s 2.48s
72. The Bad • Unoptimized patterns often have terrible performance characteristics • Numeric operations allocate objects • Instance vars require wrapper, indirection • Java objects must be wrapped or marshaled • Closure state is expensive • Blah blah startup time blah blah
73. Performance Futures
74. Intermediate Representation • Traditional compiler architecture for JRuby • Instructions, operands, control-flow graph • Optimization passes • Optimize before emitting bytecode • Closer mapping to JVM operations
75. def foo(a, b) c = 1 d = a + c end 0 check_arity(2, 0, -1) 1 a = recv_pre_reqd_arg(0) 2 b = recv_pre_reqd_arg(1) 3 %block = recv_closure 4 thread_poll 5 line_num(1) 6 c = 1 7 line_num(2) 8 %v_0 = call(:+, a, [c]) 9 d = copy(%v_0) 10 return(%v_0) Register-based 3 address format IR Instructions Semantic Analysis
76. The Good • Well-understood techniques • Creator worked on Java JIT years ago • Drastically simpler JVM bytecode backend • Less one-off opto code • Simpler runtime • Performance!
77. Numeric loop performance 0 0.9 1.8 2.7 3.6 times faster than MRI 2.1 JRuby 1.7 JRuby 9k
78. def loop(a)
 i = 0
 while i < a
 i +=1
 end
 end def loop(a:int)
 i = 0
 while i < a
 i +=1
 end
 end
79. Numeric loop performance 0 15 30 45 60 times faster than MRI 2.1 JRuby 9k JRuby 1.7 9k+unbox
80. The Bad • Work in progress - join us! • Large-scale rework • Requires knowledge of compiler design • Doesn’t address object layout yet
81. Truffle • Optimizing language framework • Builds on Graal • Write AST + interpreter, it builds a JIT • Hints for object layout, unboxing • Early days but very exciting
82. mandelbrot(500) 0 10 20 30 40 times faster than MRI 2.1 JRuby 9k + indy JRuby 9k + unboxing JRuby 9k + Truffle
83. GC
84. Time per GC versus heap usage Time per GC 0ms 75ms 150ms 225ms 300ms Heap usage (MRI/JRuby) 188KB/29MB Ruby 2.0.0 JRuby
85. GC runs per iteration, gc_stress.rb 0 750 1500 2250 3000 Ruby 2.2 JRuby
86. GC time per iteration, gc_stress.rb 0 0.6 1.2 1.8 2.4 Ruby 2.2 JRuby
87. Many GCs • Serial collector • Parallel collector • Concurrent collector (CMS) • Concurrent region collector (G1) • Continuously concurrent collector (Azul)
88. JRuby, One Thread
89. JRuby, One Thread Serial GC Parallel GC Concurrent GC
90. Unlikely GC will be a problem for most apps.
91. Tooling
92. JVM tools are awesome
94. $ jruby --manage …
97. require 'jmx'
 
 def in_mb(value)
 format "%0.2f Mb" % (value.to_f / (1024 * 1024))
 end
 
 server = JMX.simple_server
 client = JMX.connect
 memory = client["java.lang:type=Memory"]
 
 Thread.new do
 puts "Enter 'gc' to garbage collect or anything else to quit"
 while gets.chomp == "gc"
 memory.gc
 end
 
 server.stop
 exit 0
 end
 
 while (true)
 heap = in_mb(memory.heap_memory_usage.used)
 non_heap = in_mb(memory.non_heap_memory_usage.used)
 
 puts "Heap: #{heap}, Non-Heap: #{non_heap}"
 sleep(2)
 end
98. require 'jmx'
 
 class UsefulMetrics < RubyDynamicMBean
 rw_attribute :name, :string, "My sample attribute"
 r_attribute :explicit_reader, :int, "Sample int with writer", :my_reader
 
 operation "Doubles a value"
 parameter :int, "a", "Value to double"
 returns :int
 def double(a)
 a + a
 end
 end
 
 my_server = JMX::MBeanServer.new
 
 dyna = UsefulMetrics.new("domain.UsefulMetrics", $$.to_s)
 domain = my_server.default_domain
 my_server.register_mbean dyna, "#{domain}:type=UsefulMetrics"
99. Heap Dumps VisualVM or jmap + jhat
100. Useful to Us
101. Maybe Useful for You
102. Tooling Futures
103. Alienist Java Heap Ruby Heap hprof JSON https://github.com/enebo/alienist_viewer https://github.com/enebo/alienist Ruby Heap JSON Rails App
104. More Stats • What metrics would be interesting? • Hook up to monitoring services? • Remote debugging and profiling?
105. alienist $ jruby -e 'gets'& $ jmap -dump:live,file=foo.dump <pid> $ alienist -c foo.dump Took 33.7012s to parse 596577 objects from 5698 classes. Ruby Instances: 11,627, Ruby Classes: 351
106. { "name": "Gem::Platform", "size": 325, "id": 34200727296, "instances": [ { "id": 34211817960, "size": 40, "variables": [ ["@cpu", 34186089840], ["@os", 34211854176], ["@version", 34186089840] ], "references": [34211803176] }, ] }, Class name ids!
107. alienist_viewer
108. alienist_viewer
109. alienist TODO • Concept of Size, query language? • Make (someone make) MRI dumper • Allow for impl-specific data • Specification for JSON format
110. Your Turn
111. Excluding merges, 24 authors have pushed 790 commits to master and 909 commits to all branches. On master, 1,880 files have changed and there have been 69,878 additions and 66,549 deletions.
112. jruby.org
113. Use 'jruby-9000' in Ruby installers
114. Test with 'jruby-head' on TravisCI
115. File bugs at bugs.jruby.org
116. Help us build Ruby’s Future!
117. Thank you!