slides

z, ?	toggle help (this)
space, →	next slide
shift-space, ←	previous slide
d	toggle debug mode
## <ret>	go to slide #
c, t	table of contents (vi)
f	toggle footer
r	reload slides
n	toggle notes
p	run preshow

loading presentation...

From
Profiling Ruby
to
Frankenstein Programming

or how I learned
to stop worrying
and love using Ruby
for scientific computations

me, me, me.

(that’s Piotr Szotkowski)
(or @chastell)

Mr Hyde

archive.org / details / DrJekyllandMrHyde

Dr Jekyll

archive.org / details / DrJekyllandMrHyde

Mr Hyde

archive.org / details / DrJekyllandMrHyde

WUT

Dr Jekyll

archive.org / details / DrJekyllandMrHyde

FORTRAN

www.theage.com.au / news / technology / programming-language-pioneer-passes-away / 2007 / 03 / 21 / 1174153117713.html

Arc

en.wikipedia.org / wiki / Paul_Graham_(computer_programmer)

LISP

www-formal.stanford.edu / jmc / frames.html

en.wikipedia.org / wiki / Dennis_Ritchie

C++

www.linuxjournal.com / article / 7099

Java

en.wikipedia.org / wiki / James_Gosling

Python

commons.wikimedia.org / wiki / Category:Guido_van_Rossum

Python Cookbook

oreilly.com / catalog / 9780596001674

T-shirt

flickr.com / john_lam / 1910968816

Matz

flickr.com / john_lam / 1910968816

Ruby’s
performance

Ruby’s
‘performance’

Extreme Slow Food

www.futures.hawaii.edu / 2008 / 09 / dator-on-slow-food.php

A Taste of Slow

slowfoodaustralia.com.au / events / national / a-taste-of-slow

shootout
.

alioth
.

debian
.

org

Ruby vs C

Ruby vs C¹

Ruby vs C²

Ruby vs C³

slow user

www.computerrepairmaintenance.com / computer-help / what-can-i-do-to-fix-my-slow-computer

but… but…
benchmarks?
srsly?

Synthetic benchmarks tell us sweet FA
about real world performance of code,
architecture
being a much more significant consideration than the proportion of raw MIPS a given language will deliver on a given platform.

The average netbook could happily run
all of Teller’s fusion bomb models
along with the full telemetry analysis
of all the Apollo missions
in the pauses between loading XKCD comics
and binning junk mail
without the user being any the wiser.

Eleanor McHugh

www.flickr.com / photos / gavinbell / 244169511

profiling
Ruby

Euler

en.wikipedia.org / wiki / Leonhard_Euler

Project Euler

starting in the top left corner of a 2×2 grid, there are 6 routes (without backtracking)
to the bottom right corner:
how many routes are there
through a 20×20 grid?

Euler 15

only right and down
so, let’s say… zeroes and ones?
0011, 0101, 0110, 1001, 1010, 1100
all (2+2)-digit binary numbers
with the same number of ones and zeroes



grid = 2



grid = 2

(0...2**(2 * grid))                          # (0-15)



grid = 2

(0...2**(2 * grid))                          # (0-15)
.map { |n| n.to_s 2 }                        # ['0', '1', '10', '11'…]



grid = 2

(0...2**(2 * grid))                          # (0-15)
.map { |n| n.to_s 2 }                        # ['0', '1', '10', '11'…]
.map { |s| s.rjust 2 * grid, '0' }           # ['0000','0001','0010'…]



grid = 2

(0...2**(2 * grid))                          # (0-15)
.map { |n| n.to_s 2 }                        # ['0', '1', '10', '11'…]
.map { |s| s.rjust 2 * grid, '0' }           # ['0000','0001','0010'…]
.select { |s| s.count('1') == s.count('0') } # ['0011','0101','0110'…]



grid = 2

(0...2**(2 * grid))                          # (0-15)
.map { |n| n.to_s 2 }                        # ['0', '1', '10', '11'…]
.map { |s| s.rjust 2 * grid, '0' }           # ['0000','0001','0010'…]
.select { |s| s.count('1') == s.count('0') } # ['0011','0101','0110'…]
.size                                        # 6

# ruby grid.rb <number>

grid = ARGV.first.to_i

puts (0...2**(2 * grid))
.map { |n| n.to_s 2 }
.map { |s| s.rjust 2 * grid, '0' }
.select { |s| s.count('1') == s.count('0') }
.size

$ ruby grid.rb 2
6

$ ruby grid.rb 20
^Cgrid.rb:2:in `block in <main>': Interrupt
  from grid.rb:2:in `each'
  from grid.rb:2:in `map'
  from grid.rb:2:in `<main>'

$ time ruby grid.rb 2
6
real 0m0.017s
user 0m0.020s
sys  0m0.000s

$ time ruby grid.rb 8
12870
real 0m0.204s
user 0m0.180s
sys  0m0.020s

$ time ruby grid.rb 10
184756
real 0m3.040s
user 0m2.890s
sys  0m0.130s

$ time ruby grid.rb 11
705432
real 0m13.291s
user 0m12.750s
sys  0m0.490s

Before microscopes

icanhascheezburger.com / 2011 / 03 / 18 / funny-pictures-before-microscopes

`require 'profile'`

$ time ruby grid.rb 9
48620
real 0m0.827s
user 0m0.780s
sys  0m0.040s

$ time ruby -r profile grid.rb 9
48620
  %   cumulative   self              self     total
 time   seconds   seconds    calls  ms/call  ms/call  name
 32.38    12.64     12.64        1 12640.00 18570.00  Array#select
 23.64    21.87      9.23        1  9230.00 13610.00  Array#map
 12.09    26.59      4.72        1  4720.00  6840.00  Range#each
 10.04    30.51      3.92   524288     0.01     0.01  String#count
  6.25    32.95      2.44   262144     0.01     0.01  String#rjust
  5.43    35.07      2.12   262145     0.01     0.01  Fixnum#to_s
  5.15    37.08      2.01   262144     0.01     0.01  Fixnum#==
  4.97    39.02      1.94   262145     0.01     0.01  Fixnum#*
  0.05    39.04      0.02        2    10.00    10.00  IO#write
  0.00    39.04      0.00        2     0.00     0.00  IO#set_encoding
  0.00    39.04      0.00        1     0.00  6840.00  Enumerable.map
  0.00    39.04      0.00        1     0.00     0.00  Fixnum#**
  0.00    39.04      0.00        1     0.00     0.00  String#to_i
  0.00    39.04      0.00        1     0.00     0.00  Array#first
  0.00    39.04      0.00        1     0.00     0.00  Array#size
  0.00    39.04      0.00       28     0.00     0.00  Hash#default
  0.00    39.04      0.00        1     0.00    20.00  IO#puts
  0.00    39.04      0.00        1     0.00    20.00  Kernel.puts
  0.00    39.04      0.00        1     0.00 39040.00  #toplevel
real 0m40.984s
user 0m39.080s
sys  0m1.260s

Sun racks

www.anchor.com.au / blog / 2009 / 10 / ooh-bugger

gem install ruby-prof

$ time ruby-prof grid.rb 9
48620
Thread ID: 14590720
Total: 3.150000

 %self     total     self     wait    child    calls  name
 22.86      1.30     0.72     0.00     0.58        1  Array#select
 18.41      0.58     0.58     0.00     0.00   524288  String#count
 17.14      0.89     0.54     0.00     0.35        1  Range#each
 15.87      0.96     0.50     0.00     0.46        1  Array#map
 14.60      0.46     0.46     0.00     0.00   262144  String#rjust
 11.11      0.35     0.35     0.00     0.00   262145  Fixnum#to_s
  0.00      0.00     0.00     0.00     0.00        2  IO#write
  0.00      0.00     0.00     0.00     0.00        1  IO#puts
  0.00      0.00     0.00     0.00     0.00        1  Kernel#puts
  0.00      0.89     0.00     0.00     0.89        1  Enumerable#map
  0.00      0.00     0.00     0.00     0.00        1  <Class::Range>#allocate
  0.00      0.00     0.00     0.00     0.00        1  Fixnum#**
  0.00      0.00     0.00     0.00     0.00        1  String#to_i
  0.00      0.00     0.00     0.00     0.00        1  Array#first
  0.00      0.00     0.00     0.00     0.00       28  Hash#default
  0.00      0.00     0.00     0.00     0.00        2  IO#set_encoding
  0.00      3.15     0.00     0.00     3.15        1  Kernel#load
  0.00      3.15     0.00     0.00     3.15        2  Global#[No method]

real 0m3.425s
user 0m2.560s
sys  0m0.830s

$ ruby-prof -p graph_html -f grid.html grid.rb 9
48620

grid.html.png

slow goats

www.anchor.com.au / blog / 2009 / 10 / ooh-bugger

gem install perftools.rb

$ export CPUPROFILE=grid

$ time ruby -r perftools grid.rb 9
48620
PROFILE: interrupts/evictions/bytes = 76/0/272

real 0m0.928s
user 0m0.830s
sys  0m0.090s

$ pprof.rb --text grid
Using local file /home/chastell/.rvm/rubies/ruby-1.9.2-p180/bin/ruby.
Using local file grid.
Total: 76 samples
      16  21.1%  21.1%       16  21.1% garbage_collector
      12  15.8%  36.8%       12  15.8% String#count
      11  14.5%  51.3%       11  14.5% Fixnum#to_s
      10  13.2%  64.5%       22  28.9% Array#select
      10  13.2%  77.6%       21  27.6% Range#each
      10  13.2%  90.8%       10  13.2% String#rjust
       7   9.2% 100.0%       17  22.4% Array#map
       0   0.0% 100.0%       21  27.6% Enumerable#map

$ time ruby grid.rb 10
184756

real 0m3.082s
user 0m2.910s
sys  0m0.150s

$ time ruby -r perftools grid.rb 10
184756
PROFILE: interrupts/evictions/bytes = 293/0/272

real 0m3.189s
user 0m2.940s
sys  0m0.230s

$ pprof.rb --text grid
Using local file /home/chastell/.rvm/rubies/ruby-1.9.2-p180/bin/ruby.
Using local file grid.
Total: 293 samples
      67  22.9%  22.9%       67  22.9% Fixnum#to_s
      56  19.1%  42.0%       78  26.6% Array#select
      45  15.4%  57.3%       45  15.4% garbage_collector
      44  15.0%  72.4%       72  24.6% Array#map
      31  10.6%  82.9%       98  33.4% Range#each
      28   9.6%  92.5%       28   9.6% String#rjust
      22   7.5% 100.0%       22   7.5% String#count
       0   0.0% 100.0%       98  33.4% Enumerable#map

$ pprof.rb --pdf grid > grid.pdf
Using local file /home/chastell/.rvm/rubies/ruby-1.9.2-p180/bin/ruby.
Using local file grid.
Dropping nodes with <= 1 samples; edges with <= 0 abs(samples)

`$ ruby -r perftools grid.rb 10`

grid 10

`$ ruby -r perftools grid.rb 11`

grid 11

`$ ruby -r perftools grid.rb 12`

grid 12

grid = ARGV.first.to_i

puts (0...2**(2 * grid))
.map { |n| n.to_s 2 }
.map { |s| s.rjust 2 * grid, '0' }
.select { |s| s.count('1') == s.count('0') }
.size

grid = ARGV.first.to_i

puts (0...2**(2 * grid))
.map { |n| n.to_s 2 }
.select { |s| s.count('1') == grid }
.size

grid = ARGV.first.to_i

puts (0...2**(2 * grid))
.select { |n| n.to_s(2).count('1') == grid }
.size

grid = ARGV.first.to_i

puts (0...2**(2 * grid))
.count { |n| n.to_s(2).count('1') == grid }

grid 12 old

$ time ruby -r perftools grid.rb 12
2704156
PROFILE: interrupts/evictions/bytes = 5640/0/272

real 0m57.870s
user 0m55.000s
sys  0m2.690s

grid 12 new

$ time ruby -r perftools grid-short.rb 12
2704156
PROFILE: interrupts/evictions/bytes = 2744/0/200

real 0m27.676s
user 0m27.420s
sys  0m0.130s

Westheimer’s Discovery

a couple of months in the laboratory
can frequently save
a couple of hours in the library

passports

sfwchan.com / ?id=4823

I like big books

www.etsy.com / listing / 62235991 / i-like-big-books-canvas-tote-bag

class Integer
  def popcount_to_s
    to_s(2).count '1'
  end
end

grid = ARGV.first.to_i
puts (0...2**(2 * grid)).count { |n| n.popcount_to_s == grid }









$ time ruby grid-classy.rb 12
2704156
real 0m26.562s
user 0m26.370s
sys  0m0.100s

class Integer
  def popcount_bitshift
    count = 0
    number = self
    while number > 0
      number &= number - 1
      count += 1
    end
    count
  end
end

grid = ARGV.first.to_i
puts (0...2**(2 * grid)).count { |n| n.popcount_bitshift == grid }

class Integer
  def popcount_bitshift
    count = 0
    number = self
    while number > 0
      number &= number - 1
      count += 1
    end
    count
  end
end

grid = ARGV.first.to_i
puts (0...2**(2 * grid)).count { |n| n.popcount_bitshift == grid }



$ time ruby grid-bitshift.rb 12
2704156
real 0m33.155s
user 0m33.010s
sys  0m0.010s

grid bitshift

require 'inline'
class Integer
  inline do |builder|
    builder.c '
      int popcount_bitshift_c() {
        long number = NUM2LONG(self);
        int count;
        for (count = 0; number; count++) number &= number - 1;
        return count;
      }
    '
  end
end

grid = ARGV.first.to_i
puts (0...2**(2 * grid)).count { |n| n.popcount_bitshift_c == grid }

require 'inline'
class Integer
  inline do |builder|
    builder.c '
      int popcount_bitshift_c() {
        long number = NUM2LONG(self);
        int count;
        for (count = 0; number; count++) number &= number - 1;
        return count;
      }
    '
  end
end

grid = ARGV.first.to_i
puts (0...2**(2 * grid)).count { |n| n.popcount_bitshift_c == grid }

$ time ruby grid-bitshift-c.rb 12
2704156
real 0m4.649s
user 0m4.590s
sys  0m0.040s

require 'inline'
class Integer
  inline do |builder|
    builder.c '
      int popcount_builtin() {
        return __builtin_popcountl(NUM2LONG(self));
      }
    '
  end
end

grid = ARGV.first.to_i
puts (0...2**(2 * grid)).count { |n| n.popcount_builtin == grid }




$ time ruby grid-builtin.rb 12
2704156
real 0m4.559s
user 0m4.390s
sys  0m0.110s

grid builtin

Westheimer’s Discovery

a couple of months in the laboratory
can frequently save
a couple of hours in the library

LSD and You

www.weakstream.us / ?attachment_id=8226

we want to make 2 × grid steps
by combining grid right and grid down ones
pick grid elements from a set of 2 × grid elements
number of such combinations is known:
it’s the binomial coefficient of 2 × grid over grid
it’s equal to (2 × grid)! / (grid! × (2 × grid - grid)!)
for grid = 2 it’s 4! / (2! × (4 - 2)!)
for grid = 20 it’s 40! / (20! × (40 - 20)!)
(40 × 39 × … × 21) / (20 × 19 × … × 1) = 137,846,528,820

benchmarking

Project Euler

the sum of the primes below 10
is 2 + 3 + 5 + 7 = 17
find the sum of all the primes
below two million

$on math$

drugs.soup.io / post / 54096687 / This-is-not-normal-but-in-math

class Integer

  def prime?
    # …
  end

end

puts (2...2_000_000).select(&:prime?).inject :+

class Integer

  def stupid_prime?
    prime = true
    (2...self).each { |i| prime = false if (self % i).zero? }
    prime
  end

end

class Integer

  def tedious_prime?
    (2...self).all? { |i| (self % i).nonzero? }
  end

end

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end

end

cool story, bro, but…
require 'prime'
gives us
Integer#prime?
like, already.

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end

end

require 'benchmark'
require 'prime'
require './euler-primes'

Benchmark.bmbm do |bench|
  bench.report 'stupid:' do
    (2..10_000).each { |i| i.stupid_prime? }
  end
  bench.report 'tedious:' do
    (2..10_000).each { |i| i.tedious_prime? }
  end
  bench.report 'simple:' do
    (2..10_000).each { |i| i.simple_prime? }
  end
  bench.report 'clever:' do
    (2..10_000).each { |i| i.clever_prime? }
  end
  bench.report 'stdlib:' do
    (2..10_000).each { |i| i.prime? }
  end
end

$ ruby benchmark.rb
Rehearsal --------------------------------------------
stupid:   13.080000   0.000000  13.080000 ( 13.107838)
tedious:   2.090000   0.000000   2.090000 (  2.092263)
simple:    0.070000   0.000000   0.070000 (  0.069488)
clever:    0.040000   0.000000   0.040000 (  0.043665)
stdlib:    0.150000   0.000000   0.150000 (  0.145920)
---------------------------------- total: 15.430000sec

               user     system      total        real
stupid:   13.030000   0.000000  13.030000 ( 13.061577)
tedious:   2.080000   0.000000   2.080000 (  2.089508)
simple:    0.070000   0.000000   0.070000 (  0.067561)
clever:    0.040000   0.000000   0.040000 (  0.038180)
stdlib:    0.140000   0.000000   0.140000 (  0.145690)

require 'benchmark'
require 'prime'
require './euler-primes'

Benchmark.bmbm do |bench|
  bench.report 'simple:' do
    (2..100_000).each { |i| i.simple_prime? }
  end
  bench.report 'clever:' do
    (2..100_000).each { |i| i.clever_prime? }
  end
  bench.report 'stdlib:' do
    (2..100_000).each { |i| i.prime? }
  end
end

$ ruby benchmark-stdlib.rb
Rehearsal -------------------------------------------
simple:   1.190000   0.000000   1.190000 (  1.195284)
clever:   0.660000   0.010000   0.670000 (  0.664936)
stdlib:   2.210000   0.000000   2.210000 (  2.220644)
---------------------------------- total: 4.070000sec

              user     system      total        real
simple:   1.180000   0.000000   1.180000 (  1.184010)
clever:   0.660000   0.000000   0.660000 (  0.661777)
stdlib:   2.210000   0.000000   2.210000 (  2.221904)

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end







end

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end

  # 173.simple_prime? goes from 2 to 13





end

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end

  # 173.simple_prime? goes from 2 to 13
  # …checking (among others) divisibility by 2, 4, 6, 8, 10, 12…




end

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end

  def clever_prime?
    return true  if self == 2
    return false if self.even?
    3.step(Math.sqrt(self).floor, 2).all? { |i| (self % i).nonzero? }
  end

end

algorithms are hard
– let’s go caching

(…and now you have two problems)

profiling web applications

Rack-based for starters
(Rails is Rack-based)
Sinatra app + Rack server
tributary: a tiny blogging engine (yay!)
let’s ApacheBench an install with many items

$ export CPUPROFILE=profile

$ rackup -r perftools
[2010-10-21 14:59:39] INFO  WEBrick 1.3.1
[2010-10-21 14:59:39] INFO  ruby 1.9.2 (2010-08-18) [x86_64-linux]
[2010-10-21 14:59:39] INFO  WEBrick::HTTPServer#start: pid=346 port=9292
[…]

$ ab -n 10 http://localhost:9292/
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
[…]
Time taken for tests:   66.078 seconds
[…]
Requests per second:    0.15 [#/sec] (mean)
Time per request:       6607.803 [ms] (mean)
Time per request:       6607.803 [ms] (mean, across all concurrent requests)
Transfer rate:          0.41 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       0
Processing:  6536 6608 105.3   6581    6879
Waiting:     6535 6607 105.3   6581    6878
Total:       6536 6608 105.4   6581    6879
[…]

$ pprof.rb --pdf profile > profile.pdf
Dropping nodes with <= 33 samples; edges with <= 6 abs(samples)

no cache

no cache detail

module Tributary class Item < OpenStruct

  def date
    case @table[:date]
    when Date   then @table[:date].to_time
    when String then Time.parse @table[:date]
    when Time   then @table[:date]
    end
  end

end end

module Tributary class Item < OpenStruct

  def date
    @date ||= case @table[:date]
              when Date   then @table[:date].to_time
              when String then Time.parse @table[:date]
              when Time   then @table[:date]
              end
  end

end end

$ ab -n 10 http://localhost:9292/
[…]
Time taken for tests:   18.525 seconds
[…]
Requests per second:    0.54 [#/sec] (mean)
Time per request:       1852.549 [ms] (mean)
Time per request:       1852.549 [ms] (mean, across all concurrent requests)
Transfer rate:          1.46 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       0
Processing:  1797 1852 102.2   1819    2140
Waiting:     1797 1852 102.2   1819    2140
Total:       1797 1852 102.2   1819    2140

Percentage of the requests served within a certain time (ms)
  50%   1819
  66%   1837
  75%   1840
  80%   1846
  90%   2140
  95%   2140
  98%   2140
  99%   2140
 100%   2140 (longest request)

$ pprof.rb --pdf profile > profile.pdf
Dropping nodes with <= 9 samples; edges with <= 1 abs(samples)

date

date detail

module Tributary class Item < OpenStruct

  def path
    File.basename(@file, '.md').split('.').first
  end

end end

module Tributary class Item < OpenStruct

  def path
    @path ||= File.basename(@file, '.md').split('.').first
  end

end end

$ ab -n 10 http://localhost:9292/
[…]
Time taken for tests:   4.298 seconds
[…]
Requests per second:    2.33 [#/sec] (mean)
Time per request:       429.797 [ms] (mean)
Time per request:       429.797 [ms] (mean, across all concurrent requests)
Transfer rate:          6.28 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       0
Processing:   390  430 103.3    401     723
Waiting:      390  429 103.3    400     723
Total:        390  430 103.4    401     724

Percentage of the requests served within a certain time (ms)
  50%    401
  66%    401
  75%    403
  80%    403
  90%    724
  95%    724
  98%    724
  99%    724
 100%    724 (longest request)

$ pprof.rb --pdf profile > profile.pdf
Dropping nodes with <= 9 samples; edges with <= 1 abs(samples)

path

path detail

but what if we actually
cached
the whole item stream…?

$ ab -n 10 http://localhost:9292/
[…]
Time taken for tests:   0.919 seconds
[…]
Requests per second:    10.88 [#/sec] (mean)
Time per request:       91.883 [ms] (mean)
Time per request:       91.883 [ms] (mean, across all concurrent requests)
Transfer rate:          29.37 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.2      0       1
Processing:    22   92 191.9     37     637
Waiting:       22   91 191.9     36     637
Total:         22   92 192.1     37     638

Percentage of the requests served within a certain time (ms)
  50%     37
  66%     39
  75%     39
  80%     45
  90%    638
  95%    638
  98%    638
  99%    638
 100%    638 (longest request)

…and un-cached
Item#date and Item#path?

$ ab -n 10 http://localhost:9292/
[…]
Time taken for tests:   7.116 seconds
[…]
Requests per second:    1.41 [#/sec] (mean)
Time per request:       711.577 [ms] (mean)
Time per request:       711.577 [ms] (mean, across all concurrent requests)
Transfer rate:          3.79 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.1      0       0
Processing:    31  711 2119.9     44    6745
Waiting:       30  711 2119.9     43    6745
Total:         31  712 2120.0     44    6745

Percentage of the requests served within a certain time (ms)
  50%     44
  66%     44
  75%     44
  80%     44
  90%   6745
  95%   6745
  98%   6745
  99%   6745
 100%   6745 (longest request)

||
||
||
speeding
Ruby
up
˙

alternative
implementations

JRuby
MacRuby
Rubinius

alternative

fast
libraries

NArray
Ruby/GSL
ruby-boost-regex
google_hash

Google hash

gpsobsessed.com / cops-stumble-upon-a-2-acre-marijuana-field-in-google-earth

Strangelove
programming

Teller

en.wikipedia.org / wiki / Edward_Teller

Strangelove

worldpoliticsblog.wordpress.com / 2010 / 07 / 14 / paging-dr-strangelove

Ruby/DL

&

Ruby FFI

gem install levenshtein

Levenshtein distance between
two Strings, Arrays, etc.
Strings, Arrays and Arrays
of Strings: optimized C
everything else: C
if C not available: Ruby

bridges

C via MRI
D via RuDy
Haskell via Hubris
Java via JRuby
R via Rserve

Frankenstein
programming

what about embedding
foreign language?

(after all, you can do Assembler in C…)
Yes We Can:
RubyInline
JavaInline
Hubris

require 'inline'

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end














end

require 'inline'

class Integer

  def simple_prime?
    (2..Math.sqrt(self).floor).all? { |i| (self % i).nonzero? }
  end

  inline do |builder|
    builder.c '
      VALUE inline_simple() {
        long number = NUM2LONG(self);
        long i;
        for (i = 2; i <= floor(sqrt(number)); i++) {
          if (number % i == 0) return Qfalse;
        }
        return Qtrue;
      }
    '
  end

end

require 'inline'

class Integer

  inline do |builder|
    builder.c '
      VALUE inline_clever() {
        long number = NUM2LONG(self);
        if (number == 2)     return Qtrue;
        if (number % 2 == 0) return Qfalse;
        long i;
        for (i = 3; i <= floor(sqrt(number)); i = i + 2) {
          if (number % i == 0) return Qfalse;
        }
        return Qtrue;
      }
    '
  end

end

require 'benchmark'
require 'prime'
require './euler-primes'

Benchmark.bmbm do |bench|
  bench.report 'simple:' do
    (2..100_000).each { |i| i.simple_prime? }
  end
  bench.report 'clever:' do
    (2..100_000).each { |i| i.clever_prime? }
  end
  bench.report 'stdlib:' do
    (2..100_000).each { |i| i.prime? }
  end
  bench.report 'inline:' do
    (2..100_000).each { |i| i.inline_clever }
  end
end

$ ruby benchmark-inline.rb
Rehearsal -------------------------------------------
simple:   1.180000   0.000000   1.180000 (  1.187370)
clever:   0.670000   0.000000   0.670000 (  0.668914)
stdlib:   2.300000   0.000000   2.300000 (  2.306439)
inline:   0.070000   0.000000   0.070000 (  0.075871)
---------------------------------- total: 4.220000sec

              user     system      total        real
simple:   1.170000   0.000000   1.170000 (  1.177349)
clever:   0.680000   0.000000   0.680000 (  0.673092)
stdlib:   2.290000   0.010000   2.300000 (  2.303018)
inline:   0.070000   0.000000   0.070000 (  0.077858)

the right solution?

require 'prime'

puts Prime.take_while { |p| p < 2_000_000 }.inject :+

world domination plans

first profile
then optimise algorithms
(cache? parallelise? benchmark!)
rewrite last
(using the right tool)
search: suprisingly many common problems
are actually already solved…

whither else?

automagic Ruby-to-C translators
ruby-optimization mailing list
threads and processes
DRb and BrB (DRb on EM)
memprof, CUDA, Thrust, Coal…

but why not write optimized code right from the start?

well, because it’s premature optimisation!

We should forget
about small efficiencies,
say about 97%
of the time:
premature optimization
is the root of all evil.

Donald Knuth

laughingsquid.com / jacob-appelbaum-donald-knuth-demonstrate-the-recursive-homeboys-principle

homeboy

laughingsquid.com / jacob-appelbaum-donald-knuth-demonstrate-the-recursive-homeboys-principle

homeboy¹

laughingsquid.com / jacob-appelbaum-donald-knuth-demonstrate-the-recursive-homeboys-principle

homeboy²

laughingsquid.com / jacob-appelbaum-donald-knuth-demonstrate-the-recursive-homeboys-principle

questions?

chastell @ chastell.net

github.com / chastell

	@chastell
	@chastell

talks.chastell.net

links

IM Fell English & Inconsolata fonts:	code.google.com / webfonts
CiviCRM:	civicrm.org
Computer Language Benchmarks Game:	shootout.alioth.debian.org
Ellen McHugh quote:	ruby-talk:355820
ruby-prof:	github.com / rdp / ruby-prof
perftools.rb:	github.com / tmm1 / perftools.rb
Project Euler:	projecteuler.net
JRuby:	jruby.org
MacRuby:	macruby.org
Rubinius:	rubini.us
Ruby VM Shootout:	mitkokostov.info / post / 1108719323 / ruby-vm-shootout
NArray:	narray.rubyforge.org
Ruby/GSL:	rb-gsl.rubyforge.org
ruby-boost-regex:	github.com / michaeledgar / ruby-boost-regex
google_hash:	github.com / rdp / google_hash
levenshtein:	rubyforge.org / projects / levenshtein

links (cont.)

RuDy:	github.com / tomash / rudy
Hubris:	github.com / mwotton / Hubris
Rserve Ruby client:	github.com / clbustos / Rserve-Ruby-client
RubyInline:	zenspider.com / ZSS / Products / RubyInline
JavaInline:	kenai.com / projects / java-inline
Rob Pike’s Public Static Void:	www.youtube.com / watch?v=5kj5ApnhPAE
ruby2c:	rubyforge.org / projects / ruby2c
crystalizer:	github.com / rdp / crystalizer
ruby-optimization ML:	groups.google.com / group / ruby-optimization
DRb:	segment7.net/projects / ruby / drb
BrB:	www.tricksonrails.com / 2010 / 04 / introducing-brb-extremely-fast-interface-for-doing-distributed-ruby
memprof:	github.com / ice799 / memprof
Ruby and CUDA and Thrust, Oh My!:	coreyhoffstein.com / 2009 / 06 / 15 / ruby-and-cuda-and-thrust-oh-my
Coal:	dismaldenizen.github.com / coal
Donald Knuth quote:	Structured Programming with go to Statements
Geekz Shop:	geekz.co.uk / shop

FromProfiling RubytoFrankenstein Programming

FromProfiling RubytoFrankenstein Programming

or how I learnedto stop worryingand love using Rubyfor scientific computations

me, me, me.

Ruby’sperformance

Ruby’s‘performance’

but… but…benchmarks?

srsly?

profilingRuby

Project Euler

require 'profile'

gem install ruby-prof

gem install perftools.rb

$ ruby -r perftools grid.rb 10

$ ruby -r perftools grid.rb 11

$ ruby -r perftools grid.rb 12

Westheimer’s Discovery

a couple of months in the laboratorycan frequently savea couple of hours in the library

Westheimer’s Discovery

a couple of months in the laboratorycan frequently savea couple of hours in the library

benchmarking

Project Euler

algorithms are hard– let’s go caching

profiling web applications

but what if we actuallycachedthe whole item stream…?

…and un-cachedItem#date and Item#path?

||||||speedingRubyup˙

alternativeimplementations

fastlibraries

Strangeloveprogramming

Ruby/DL

&

Ruby FFI

gem install levenshtein

bridges

Frankensteinprogramming

what about embeddingforeign language?

the right solution?

the right solution?

world domination plans

whither else?

but why not write optimized code right from the start?

questions?

chastell @ chastell.net

github.com / chastell

talks.chastell.net

links

links (cont.)

From
Profiling Ruby
to
Frankenstein Programming

From
Profiling Ruby
to
Frankenstein Programming

or how I learned
to stop worrying
and love using Ruby
for scientific computations

Ruby’s
performance

Ruby’s
‘performance’

but… but…
benchmarks?

profiling
Ruby

`require 'profile'`

`$ ruby -r perftools grid.rb 10`

`$ ruby -r perftools grid.rb 11`

`$ ruby -r perftools grid.rb 12`

a couple of months in the laboratory
can frequently save
a couple of hours in the library

a couple of months in the laboratory
can frequently save
a couple of hours in the library

algorithms are hard
– let’s go caching

but what if we actually
cached
the whole item stream…?

…and un-cached
Item#date and Item#path?

||
||
||
speeding
Ruby
up
˙

alternative
implementations

fast
libraries

Strangelove
programming

Frankenstein
programming

what about embedding
foreign language?