RubyConf 2009

RubiniusThe 4th Year

Tuesday, November 24, 2009

RubiniusThe 1st Senior Year


4th Year


Year 1

48 commits

Freshman Year


http://photos.dzki.co.uk/oak_sapling_warming_itself_in_the_morning_sun.jpgTuesday, November 24, 2009

http://photos.dzki.co.uk/oak_sapling_warming_itself_in_the_morning_sun.jpg

http://photos.dzki.co.uk/oak_sapling_warming_itself_in_the_morning_sun.jpg

A Toy

Year 1


An Idea

Year 1


Year 2

2802 commits

Sophomore Year


Hubris

Year 3


Hubris

1.0 anyone?

Year 3


Year 3

6172 commits

Junior Year


Compliance

Year 3


Compliance

RubySpec

Year 3


Redirection

Year 3


Redirection

C++

Year 3


Plotting

Year 3


Year 4

2220 commits

1st Senior Year


JIT

Year 4


JIT

LLVM

Year 4


JIT

Hotspot Detection

Year 4


JIT

Type Profiling

Year 4


JIT

Method Inlining

Year 4


FTW

Block Inlining!

Year 4


RubySpec

Year 4


RubySpec

93% Pass Rate

Year 4


RubySpec

rake, rspec, rails

Year 4


C Extensions

Year 4


C Extensions

syck, nokogiri, mysql, yajl-json

Year 4


Compliance+

Performance


93+% RubySpec

Compliance


93+% RubySpec

Passing/untagging more everyday

Compliance


NotImplementedError


ObjectSpace

NotImplementedError


ObjectSpace

_id2ref

NotImplementedError


ObjectSpace

each_object (limited)

NotImplementedError


callcc

NotImplementedError


callcc

Yet.

NotImplementedError


C-API

Compliance


C-API

syck, nokogiri, etc.

Compliance


C-API

A few gotchas...

Compliance


C ExtensionDo’s and Don’ts


prefer a function over an R*() macro

Do


RBASIC(obj)->class

rb_obj_class(obj)

Rubinius


rb_ivar_[get|set](obj, ...)

1.9 / Rubinius

ROBJECT(obj)->iv_tbl


use the “re.h” header

1.9 / Rubinius

Don’t


use the “env.h” header

Don’t

ruby_frameruby_scoperuby_class

ruby_dyna_varsruby_in_eval

1.9 / Rubinius


Try your gem


Report your problems!


http://github.com/evanphx/rubinius/

issues


http://github.com/evanphx/issues






Performance


The Two Fronts

Performance


Why We Fight

What We Write

Performance


(0..10).to_a.each do |i|ary << i

end


0.upto(10) do |i|ary << i

end


The Two Fronts

How We Run It

Performance


Benchmarking

Performance


Benchmarking

Expectations vs. Reality

Performance


Benchmarking

Lies, Damned Lies, and Statistics

Performance


Benchmarks


0

14

29

43

57

71

86

100

114

ruby 1.9 jruby macruby rubinius

Times faster than 1.8


0

7

14

21

29

36

43

50

57




impl

ms

x faster

1.8 1.9 jruby macruby rbx

309304 51113 113824 70580 24232

1.0 6.0 2.7 4.3 12.7


12x faster!

Benchmarks


12x faster!

On these benchmarks!

Benchmarks


12x faster!

In general?

Benchmarks


What did we benchmark?

Benchmarks


Calling a method.Empty while loop.Creating a block.Accessing ivars.

Performing ‘not’.Calling a polymorphic method.

Calling super.


Why these things?

Benchmarks


Why these things?

How do these results translate?

Benchmarks


Analogy


Dishonest?


Ruby performancebegets

core performance

Rubinius


Ruby performancebegets

?

1.8


Need more/better data

Benchmarks


def foo() ary = [] 100.times { |i| ary << i }end

300,000 times


0

2

4

6

8

1.8 1.9 macruby jruby rbx

3.26

2.60

4.243.91

7.82

Seconds


def foo() hsh = {} 100.times { |i| hsh[i] = 0 }end

100,000 times


0

2.75

5.5

8.25

11


10.5

1.67

7.62

4.74.77

Seconds


def foo() hsh = { 47 => true } 100.times { |i| hsh[i] }end

100,000 times


0

1

2

3

4


2.6

1.75

2.61

2.12

3.64

Seconds


Conclusion

Benchmarking


Conclusion

Rubinius is awesome at ruby code

Benchmarking


Conclusion

Comparing C/Java to Ruby

Benchmarking


More Ruby


Burden of our own devising

More Ruby


Fast compliance

More Ruby


Slower core performance

More Ruby


Slower core performance

Ruby vs. C/Java

More Ruby


The upside...

More Ruby


0

14

29

43

57

71

86

100

114




0

14

29

43

57

71

86

100

114




def m; nil; enddef Bench.run i=0 while i < 15_000_000 i+=1 m; m; m; m; m; m; m; m endend


impl

ms

x faster


21361 6316 10101 3295 186

1.0 3.4 2.1 6.5 114.8


impl

ms

x faster


21361 6316 10101 3295 186

1.0 3.4 2.1 6.5 114.8


impl

ms

x faster


21361 6316 10101 3295 186

1.0 3.4 2.1 6.5 114.8


impl

ms

x faster


21361 6316 10101 3295 186

1.0 3.4 2.1 6.5114.8


impl

ms

x faster


21361 6316 10101 3295 186

1.0 3.4 2.1 6.5114.8Tuesday, November 24, 2009

impl

ms

x faster


21361 6316 10101 3295 186

1.0 3.4 2.1 6.5114.8Tuesday, November 24, 2009

Bug


right?

Bug


Bug




JIT

Method Inlining

Year 4




bin/rbx -Xjit.inline.debug 0/bm_vm2_method.rb


JIT: compiling Object#runinlining: primitive fixnum_lt into run.inlining: Object#m into run (<metaclass>)inlining: Object#m into run (<metaclass>)inlining: Object#m into run (<metaclass>)inlining: Object#m into run (<metaclass>)inlining: Object#m into run (<metaclass>)inlining: Object#m into run (<metaclass>)inlining: Object#m into run (<metaclass>)inlining: Object#m into run (<metaclass>)


def m; nil; enddef Bench.run i=0 while i < 15_000_000 i+=1nil;nil;nil;nil;nil;nil;nil;nil endend


def m; nil; enddef Bench.run i=0 while i < 15_000_000 i+=1 endend


Performance


Many programs, much faster

Performance


Many programs, a little slower

Performance


Some programs, a lot slower

Performance


Releases


0.13

Releases


0.13

JIT by default

Releases


0.13

Lots of fixes

Releases


1.0rc1Tuesday, November 24, 2009

Next Week!


Expectations


Runs rails 3

Expectations


Performance improvements

Expectations


One RC every month

Expectations


One RC every month

Until 1.0 is out.

Expectations


RubyConf 2009

Technology

tuesday

2009 performance

obj

2009 http

november 24

nokogiri

gt

2009 compliance